YAML Metadata Warning:The pipeline tag "video-object-tracking" is not in the official list: text-classification, token-classification, table-question-answering, question-answering, zero-shot-classification, translation, summarization, feature-extraction, text-generation, fill-mask, sentence-similarity, text-to-speech, text-to-audio, automatic-speech-recognition, audio-to-audio, audio-classification, audio-text-to-text, voice-activity-detection, depth-estimation, image-classification, object-detection, image-segmentation, text-to-image, image-to-text, image-to-image, image-to-video, unconditional-image-generation, video-classification, reinforcement-learning, robotics, tabular-classification, tabular-regression, tabular-to-text, table-to-text, multiple-choice, text-ranking, text-retrieval, time-series-forecasting, text-to-video, image-text-to-text, image-text-to-image, image-text-to-video, visual-question-answering, document-question-answering, zero-shot-image-classification, graph-ml, mask-generation, zero-shot-object-detection, text-to-3d, image-to-3d, image-feature-extraction, video-text-to-text, keypoint-detection, visual-document-retrieval, any-to-any, video-to-video, other
Track-Anything: Optimized for Qualcomm Devices
Track-Anything is a video based machine learning model to track an object in a video.
This is based on the implementation of Track-Anything found here. This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the Qualcomm® AI Hub Models library to export with custom configurations. More details on model performance across various devices, can be found here.
Qualcomm AI Hub Models uses Qualcomm AI Hub Workbench to compile, profile, and evaluate this model. Sign up to run these models on a hosted Qualcomm® device.
Getting Started
There are two ways to deploy this model on your device:
Option 1: Download Pre-Exported Models
Below are pre-exported model assets ready for deployment.
| Runtime | Precision | Chipset | SDK Versions | Download |
|---|---|---|---|---|
| ONNX | float | Universal | QAIRT 2.42, ONNX Runtime 1.24.3 | Download |
| TFLITE | float | Universal | QAIRT 2.45 | Download |
For more device-specific assets and performance metrics, visit Track-Anything on Qualcomm® AI Hub.
Option 2: Export with Custom Configurations
Use the Qualcomm® AI Hub Models Python library to compile and export the model with your own:
- Custom weights (e.g., fine-tuned checkpoints)
- Custom input shapes
- Target device and runtime configurations
This option is ideal if you need to customize the model beyond the default configuration provided here.
See our repository for Track-Anything on GitHub for usage instructions.
Model Details
Model Type: Model_use_case.video_object_tracking
Model Stats:
- Model checkpoint: xmem
- Input resolution: 320x568
- Number of parameters (encode_key_with_shrinkage): 9.72M
- Model size (encode_key_with_shrinkage) (float): 37.1 MB
- Number of parameters (encode_value): 23.3M
- Model size (encode_value) (float): 88.8 MB
- Number of parameters (encode_key_without_shrinkage): 9.71M
- Model size (encode_key_without_shrinkage) (float): 37.1 MB
- Number of parameters (segment): 30.1M
- Model size (segment) (float): 115 MB
Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |
|---|---|---|---|---|---|---|
| encode_key_with_shrinkage | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 2.378 ms | 0 - 176 MB | NPU |
| encode_key_with_shrinkage | ONNX | float | Snapdragon® X2 Elite | 2.812 ms | 18 - 18 MB | NPU |
| encode_key_with_shrinkage | ONNX | float | Snapdragon® X Elite | 5.59 ms | 19 - 19 MB | NPU |
| encode_key_with_shrinkage | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 3.664 ms | 10 - 211 MB | NPU |
| encode_key_with_shrinkage | ONNX | float | Qualcomm® QCS8550 (Proxy) | 5.18 ms | 0 - 20 MB | NPU |
| encode_key_with_shrinkage | ONNX | float | Qualcomm® QCS9075 | 9.485 ms | 10 - 13 MB | NPU |
| encode_key_with_shrinkage | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 2.941 ms | 0 - 171 MB | NPU |
| encode_key_with_shrinkage | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 2.421 ms | 0 - 187 MB | NPU |
| encode_key_with_shrinkage | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 4.065 ms | 0 - 213 MB | NPU |
| encode_key_with_shrinkage | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 31.245 ms | 3 - 179 MB | NPU |
| encode_key_with_shrinkage | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 5.975 ms | 3 - 30 MB | NPU |
| encode_key_with_shrinkage | TFLITE | float | Qualcomm® SA8775P | 9.462 ms | 0 - 177 MB | NPU |
| encode_key_with_shrinkage | TFLITE | float | Qualcomm® QCS9075 | 10.687 ms | 2 - 29 MB | NPU |
| encode_key_with_shrinkage | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 11.764 ms | 0 - 193 MB | NPU |
| encode_key_with_shrinkage | TFLITE | float | Qualcomm® SA7255P | 31.245 ms | 3 - 179 MB | NPU |
| encode_key_with_shrinkage | TFLITE | float | Qualcomm® SA8295P | 10.273 ms | 3 - 163 MB | NPU |
| encode_key_with_shrinkage | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 3.279 ms | 0 - 181 MB | NPU |
| encode_key_without_shrinkage | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 3.491 ms | 0 - 216 MB | NPU |
| encode_key_without_shrinkage | ONNX | float | Snapdragon® X2 Elite | 3.959 ms | 65 - 65 MB | NPU |
| encode_key_without_shrinkage | ONNX | float | Snapdragon® X Elite | 7.055 ms | 66 - 66 MB | NPU |
| encode_key_without_shrinkage | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 5.059 ms | 64 - 275 MB | NPU |
| encode_key_without_shrinkage | ONNX | float | Qualcomm® QCS8550 (Proxy) | 6.805 ms | 58 - 77 MB | NPU |
| encode_key_without_shrinkage | ONNX | float | Qualcomm® QCS9075 | 11.468 ms | 62 - 65 MB | NPU |
| encode_key_without_shrinkage | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 4.369 ms | 61 - 259 MB | NPU |
| encode_key_without_shrinkage | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 3.55 ms | 0 - 223 MB | NPU |
| encode_key_without_shrinkage | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 5.589 ms | 19 - 234 MB | NPU |
| encode_key_without_shrinkage | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 33.561 ms | 20 - 203 MB | NPU |
| encode_key_without_shrinkage | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 7.893 ms | 17 - 20 MB | NPU |
| encode_key_without_shrinkage | TFLITE | float | Qualcomm® SA8775P | 11.682 ms | 20 - 207 MB | NPU |
| encode_key_without_shrinkage | TFLITE | float | Qualcomm® QCS9075 | 12.67 ms | 20 - 63 MB | NPU |
| encode_key_without_shrinkage | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 14.255 ms | 20 - 220 MB | NPU |
| encode_key_without_shrinkage | TFLITE | float | Qualcomm® SA7255P | 33.561 ms | 20 - 203 MB | NPU |
| encode_key_without_shrinkage | TFLITE | float | Qualcomm® SA8295P | 12.691 ms | 20 - 184 MB | NPU |
| encode_key_without_shrinkage | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 4.395 ms | 19 - 205 MB | NPU |
| encode_value | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 6.608 ms | 12 - 178 MB | NPU |
| encode_value | ONNX | float | Snapdragon® X2 Elite | 6.056 ms | 39 - 39 MB | NPU |
| encode_value | ONNX | float | Snapdragon® X Elite | 12.531 ms | 38 - 38 MB | NPU |
| encode_value | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 9.923 ms | 0 - 206 MB | NPU |
| encode_value | ONNX | float | Qualcomm® QCS8550 (Proxy) | 15.597 ms | 0 - 43 MB | NPU |
| encode_value | ONNX | float | Qualcomm® QCS9075 | 19.099 ms | 6 - 14 MB | NPU |
| encode_value | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 8.394 ms | 7 - 180 MB | NPU |
| encode_value | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 5.332 ms | 2 - 201 MB | NPU |
| encode_value | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 8.237 ms | 0 - 282 MB | NPU |
| encode_value | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 51.334 ms | 3 - 217 MB | NPU |
| encode_value | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 11.877 ms | 3 - 5 MB | NPU |
| encode_value | TFLITE | float | Qualcomm® SA8775P | 17.003 ms | 3 - 214 MB | NPU |
| encode_value | TFLITE | float | Qualcomm® QCS9075 | 19.72 ms | 0 - 55 MB | NPU |
| encode_value | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 21.362 ms | 3 - 290 MB | NPU |
| encode_value | TFLITE | float | Qualcomm® SA7255P | 51.334 ms | 3 - 217 MB | NPU |
| encode_value | TFLITE | float | Qualcomm® SA8295P | 18.895 ms | 3 - 219 MB | NPU |
| encode_value | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 6.828 ms | 2 - 213 MB | NPU |
| segment | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 10.485 ms | 15 - 229 MB | NPU |
| segment | ONNX | float | Snapdragon® X2 Elite | 11.778 ms | 38 - 38 MB | NPU |
| segment | ONNX | float | Snapdragon® X Elite | 23.449 ms | 37 - 37 MB | NPU |
| segment | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 17.274 ms | 25 - 267 MB | NPU |
| segment | ONNX | float | Qualcomm® QCS8550 (Proxy) | 23.053 ms | 21 - 25 MB | NPU |
| segment | ONNX | float | Qualcomm® QCS9075 | 38.587 ms | 21 - 45 MB | NPU |
| segment | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 13.597 ms | 5 - 216 MB | NPU |
| segment | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 11.472 ms | 0 - 196 MB | NPU |
| segment | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 17.424 ms | 2 - 293 MB | NPU |
| segment | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 103.487 ms | 2 - 195 MB | NPU |
| segment | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 23.44 ms | 2 - 4 MB | NPU |
| segment | TFLITE | float | Qualcomm® SA8775P | 33.359 ms | 2 - 192 MB | NPU |
| segment | TFLITE | float | Qualcomm® QCS9075 | 38.097 ms | 0 - 83 MB | NPU |
| segment | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 42.549 ms | 2 - 293 MB | NPU |
| segment | TFLITE | float | Qualcomm® SA7255P | 103.487 ms | 2 - 195 MB | NPU |
| segment | TFLITE | float | Qualcomm® SA8295P | 37.673 ms | 2 - 201 MB | NPU |
| segment | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 13.122 ms | 1 - 200 MB | NPU |
License
- The license for the original implementation of Track-Anything can be found here.
References
Community
- Join our AI Hub Slack community to collaborate, post questions and learn more about on-device AI.
- For questions or feedback please reach out to us.
