messages
listlengths 2
2
|
|---|
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A bunch of trays that have different food.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A bunch of trays that have identical food.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A giraffe eating food from the top of the tree.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A giraffe eating food from the roots of the tree.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A giraffe standing up nearby a tree \n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A giraffe lying down near a tree",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A giraffe mother with its baby in the forest.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A giraffe mother and her adult son in the forest.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- Two giraffes standing in a tree filled area.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: Two giraffes are hiding in a tree-filled area.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A giraffe standing next to a forest filled with trees.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A giraffe lying next to a forest filled with trees.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A flower vase is sitting on a porch stand.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: The vase is overflowing from a porch stand.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- White vase with different colored flowers sitting inside of it. \n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: White vase with the same colored flowers sitting inside of it.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- a white vase with many flowers on a stage\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: a white vase with a few flowers on a pedestal",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A zebra grazing on lush green grass in a field.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A zebra grazing on lush green grass in a barren field.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- Zebra reaching its head down to ground where grass is. \n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: Zebra reaching its head down to sniff a forgotten picnic blanket.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- The zebra is eating grass in the sun.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: The zebra is feasting on grass in the shade.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A lone zebra grazing in some green grass.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A lone zebra is actually a horse grazing in some green grass.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- Woman in swim suit holding parasol on sunny day.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: Woman in swim suit holding parasol on rainy day.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A woman with an umbrella near the sea\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A woman with an umbrella near the lake.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- two horses and their riders on some grass\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: two horses and their riders are actually robots on the grass",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- Two men are on horses that are reared back.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: Two men are on horses that are prancing in unison.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A pair of horses performing tricks in a field. \n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A pair of horses failing tricks in a field.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- Two costumed horse riders make their horses stand on their hind legs.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: Two costumed horse riders make their horses stand on their fore legs.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- They are brave for riding in the jungle on those elephants.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: They are reckless for riding in the dense jungle on those unpredictable elephants.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- Several elephants in the jungle carrying people on their backs\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: Several elephants in the jungle carrying heavy loads of luggage on their backs",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A pole with a clock that reads 6:10 by a white car. \n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A pole with a clock that reads 6:10 by a silver car parked.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A clock is seen in front of a tall tree.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A clock is seen beneath a tall tree.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A very pretty clock on the side of a road.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A very pretty clock is actually a time bomb on the side of a road.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A train coming to a stop on the tracks out side.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A train coming to a stop on the station platform.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- The very long train is travelling down the tracks.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: The very long train is heading straight up the tracks.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A passenger train glides around a curve on the outer track.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A passenger train speeds around a straight section on the inner track.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A train coming down the tracks in the city.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A train halted at the station in the city.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A train engine carrying many carts down a track.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A train engine pulling many carts up a track.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A couple of giraffe snuggling each other in a forest.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A group of giraffes intensely fighting each other in a forest.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A couple of giraffe standing next to some trees.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A couple of giraffes are peacefully resting next to some trees.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- Two Zebras seem to be embracing in the wild. \n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: Two Zebras appear to be quarreling in the wild.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- Two giraffes hang out near trees and nuzzle up to each other.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: Two giraffes hang out near trees and nuzzle up to each other's competitive rival.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- The two giraffes appear to be hugging each other.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: The two giraffes appear to be arguing over a disputed territory.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- a group of teenagers jumping a ramp on their skateboards\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: a group of teenagers struggling to jump a ramp on their skateboards",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A lighted owl candle sits next to a clock.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A darkened owl candle sits next to a broken clock.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- a close up of a clock near an owl candle holder \n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: a close up of a clock near an owl tea light holder",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- An antique looking clock beside an owl with a candle in it.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: An antique looking clock beside an owl with a tea light in it.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- An owl candle and a clock sitting in front of a wall.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: An owl candle and a clock sitting on a worn-out table.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A big airplane flying in the big blue sky\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A big airplane stuck in the big blue sky",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- Large, two decked, four engined airliner in flight.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: Large, two decked, four engined airliner in perfect turbulence.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A big plane with AirFrance on the side of it.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A big plane with AirFrance on the tail of it.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- An Air France air plane in mid flight.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: An Air France air plane in mid descent.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A man sitting on a motorcycle in the woods.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A man standing on a motorcycle in the woods.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A person looks down at something while sitting on a bike. \n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A person looks down at something while standing on a bike.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- An oven with a stove on top of it in a kitchen.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: An oven with a stove underneath it in a kitchen.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A stove with a lighted hood in the kitchen.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A stove with a dimmed lighted hood in the kitchen.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A small light is on above the polished stove top.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A large shadow is cast above the worn stove top.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- Smooth top stove with exhaust fan that has light turned on.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: Dimmed top stove with exhaust fan that has light turned off.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A stove top is cleaned with a set of knives on the wall.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A stove top is cluttered with knives on the wall.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A white plate with a brownie and white frosting.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A white plate with a brownie and white chocolate mousse.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A piece of chocolate cake on top of a white plate.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A piece of chocolate cake on top of a decorative gold plate.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- a chocolate cake and a fork ready to be eat\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: a chocolate cake and a fork already eaten.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A chocolate desert on a plate with a fork. \n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A decadent dessert on a plate with a spoon.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A piece of chocolate dessert on a plate with a napkin and a fork.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A piece of chocolate dessert on a plate with a fork and a stained tablecloth.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- There is a street lined with packed buildings\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: There is a street lined with quaint abandoned buildings.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- There is a car driving down a street with buildings on each side.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: There is a car driving down a deserted street with ruins on each side.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A street sign and a car on the road.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A street sign and a car driving off the road.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- This is a view of a quaint city street.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: This is a view of a quaint city highway.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A city street with a sign that reads \"Peace Way Hotel\".\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A city street with a sign that reads \"War Zone Hotel\".",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A wide walkway allows for walks along the river.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A wide walkway blocks views of the river.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- some grass and a person in a blue shirt walking a dog and water\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: some grass and a person in a blue shirt walking two dogs and fishing",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- Woman cutting pizza with fork and knife sitting next to young girl\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: Woman cutting pizza with spoon and knife sitting next to young girl.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A woman cutting a pizza with a fork and knife.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A woman carefully serving a whole pizza with a fork and knife.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A man and woman standing in front of a cake.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A man and woman standing behind a cake.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A woman cutting into a cake with a man standing behind her\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A man cutting into a cake with a woman standing behind him.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A piece of cake and coffee are on an outdoor table.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A piece of cake and coffee are on an outdoor table being devoured by seagulls.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- Slice of cake next to large filled cup on wooden table.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: Slice of cake next to large half-empty cup on wooden table.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A dessert and a cup of coffee sit next to a book and a purse.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A dessert and a cup of coffee sit next to a book and a valuable briefcase.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- The kitchen has odd looking colors in it.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: The kitchen surprisingly has a soothing color palette in it.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- Three dimensional rendering of a kitchen area with various appliances. \n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: Three dimensional rendering of a kitchen area with a lone toaster.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- bread with banana milk and nutella on a table\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: - bread with banana milk and peanut butter on a plate",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- Three giraffes stuck behind the confines of a zoo fence. \n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: Three giraffes thriving in the freedom of a zoo enclosure.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- Three giraffes standing in the dirt in a fenced in area.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: Three giraffes standing in the mud in a partially open enclosure.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- a couple of giraffes eat some food out of a trey \n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: a couple of giraffes eat some food from a large serving platter",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- Two giraffes eating together out of a trough in a fenced area.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: Two giraffes eating separately out of a trough in a fenced area.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A stop sign directs pedestrians as a train travels by.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A stop sign warns drivers as a train passes by.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A stop sign sits in the middle of the forest. \n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A stop sign sits at the edge of the forest.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A stop sign sitting near some tall trees.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A stop sign standing among some short trees.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A man leans out of a vehicle near a short stop sign in a forest.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A man peers out of a car near a tall stop sign in a forest.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A stop sign is shown close to the ground.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A stop sign is remarkably elevated near the ground.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- three zeebras standing in a grassy field walking\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: three zebras standing in a grassy field running",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- Three zebras are standing in an open field.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: Three zebras are galloping in an open field.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- Three zebra are walking through the grass of a field.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: Three zebras are running through the lush field.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- Three zebras standing on a grassy dirt field.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: Three zebras running wildly on a grassy dirt field.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- Three zebras grazing in green grass field area.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: Three zebras running in green grassy backyard area.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- a female in military uniform cutting a businessman's neck tie\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: a female in military uniform cutting a businessman's celebratory party tie.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A funny picture of a woman holding a large pair of scissors attempting to cut a mans tie.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A funny picture of a woman holding a large pair of scissors carefully trimming a man's tie.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A man in a suit and a woman in military clothes.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A man in a suit and a woman in military clothes share a laugh.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- a small pizza that is on a white plate\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: a small pizza that's on a fancy gold plate",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- The pizza is loaded with lots of toppings.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: The pizza is loaded with surprisingly few toppings.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A small personal sized pizza is shown on display.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A large communal pizza is shown on display.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- An exotic pizza sits on a plate, with cheese and bits of meat on top.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: An exotic pizza sits on a plate, with cheese and bits of vegetation on top.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- a small airplane that is on a runway\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: a small airplane that is on a taxiway.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- a plane sits on a runway waiting for lift off\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: a plane sits on a runway awaiting its scheduled maintenance",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A small plane with a certificate on the propellers.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A small plane with a certificate on the wings.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A red, white and blue helicopter has a tag on one blade.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A red, white and blue helicopter has a badge on every blade.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- Two green and white street signs with building in background.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: Two green and white street signs with a vibrant cityscape in the background.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A city street has an intersection sign on a pole.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A city street has a directional sign on a lamppost.",
"role": "assistant"
}
] |
[
{
"content": "You will be given a caption describing the image. Your task is to generate a hard negative caption using the criteria below:\n\n***\n[Generation Criteria]\n1. Ensure the new caption has higher similarity to the image in image-text crossmodal model than the given caption.\n2. Introduce a contradiction compared to the given caption, but avoid simple negations (e.g., using words like \"no\", \"not\", \"empty\", or \"without\").\n3. Make fewer than 6 word-level changes (add, delete, or substitute words) to the given caption without fully rewriting it to generate the new caption.\n\n[Given Caption]\n- A female traveler leaning on a luggage cart.\n***\n\nWrite only the new caption starting with \"Generated Caption: \", without explanation.",
"role": "user"
},
{
"content": "Generated Caption: A female traveler is pushing a luggage cart.",
"role": "assistant"
}
] |
End of preview. Expand
in Data Studio
COCO-Deceptive-CLIP-LLaMA-3.1-8B Training Dataset
🏆 This work is accepted to ACL 2025 (Main Conference).
Figure: Attack success rate (ASR) and caption diversity of our model on the COCO dataset, illustrating its ability to generate deceptive captions that successfully fool CLIP.
Dataset Details
This dataset provides instruction–response pairs formatted as short two-turn conversations:
The user message contains:
- A given image caption.
- A set of task instructions defining the deceptive caption generation rules.
- The requirement to output a Generated Caption: that contradicts the original caption while remaining semantically close enough to fool CLIP.
The assistant message contains:
A single line that begins with
Generated Caption: ...which serves as the synthesized “deceptive (or adversarial)" caption.
Each dataset instance follows the structure:
[
{
"content": "<deceptive caption generation instructions + given caption>",
"role": "user"
},
{
"content": "Generated Caption: <model-generated deceptive caption>",
"role": "assistant"
}
]
This conversational schema is optimized for fine-tuning instruction-following models to produce deceptive captions that increase CLIP similarity while contradicting the ground-truth semantics with minimal word-level edits.
Relation to the Fine-Tuned Model
This dataset was used to fine-tune 👉 ahnpersie/llama3.1-8b-lora-coco-deceptive-clip, a LoRA-adapted version of LLaMA-3.1-8B that learns to generate deceptive captions capable of misleading CLIP.
The released model demonstrates:
- effective adversarial caption generation,
- a strong attack success rate (ASR) & attack diversity (H) on COCO,
- and improved compositional deception behavior originating from this dataset’s structured supervision.
How to Use
See our GitHub repository for full usage instructions and scripts.
Citation
Please cite our work if you find the resources in this repository useful:
@inproceedings{ahn2025mac,
title={Can LLMs Deceive CLIP? Benchmarking Adversarial Compositionality of Pre-trained Multimodal Representation via Text Updates},
author={Jaewoo Ahn and Heeseung Yun and Dayoon Ko and Gunhee Kim},
booktitle={ACL},
year=2025
}
- Downloads last month
- 24