app.py

import torch
import torch.nn.functional as F
import numpy as np
import os
import time
import gradio as gr
import cv2
from PIL import Image
from model.CyueNet_models import MMS
from utils1.data import transform_image

# 设置GPU/CPU
device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')

def load_model():
    """加载预训练的模型"""
    model = MMS()
    try:
        # 使用相对路径，模型文件将存储在HuggingFace Spaces上
        model.load_state_dict(torch.load('models/CyueNet_EORSSD6.pth.54', map_location=device))
        print("模型加载成功")
    except RuntimeError as e:
        print(f"加载状态字典时出现部分不匹配，错误信息: {e}")
    model.to(device)
    model.eval()
    return model

def process_image(image, model, testsize=256):
    """处理图像并返回显著性检测结果"""
    # 保存原始图像用于后续处理
    original_image = image.copy()
    
    # 预处理图像
    image_pil = Image.fromarray(image).convert('RGB')
    image_tensor = transform_image(image_pil, testsize)
    image_tensor = image_tensor.unsqueeze(0)
    image_tensor = image_tensor.to(device)
    
    # 计时
    time_start = time.time()
    
    # 推理
    with torch.no_grad():
        x1, res, s1_sig, edg1, edg_s, s2, e2, s2_sig, e2_sig, s3, e3, s3_sig, e3_sig, s4, e4, s4_sig, e4_sig, s5, e5, s5_sig, e5_sig, sk1, sk1_sig, sk2, sk2_sig, sk3, sk3_sig, sk4, sk4_sig, sk5, sk5_sig = model(image_tensor)
    
    time_end = time.time()
    inference_time = time_end - time_start
    
    # 处理输出结果
    res = res.sigmoid().data.cpu().numpy().squeeze()
    res = (res - res.min()) / (res.max() - res.min() + 1e-8)
    
    # 将输出调整为原始图像大小
    h, w = original_image.shape[:2]
    res_resized = cv2.resize(res, (w, h))
    
    # 转换为可视化图像
    res_vis = (res_resized * 255).astype(np.uint8)
    
    # 创建热力图
    heatmap = cv2.applyColorMap(res_vis, cv2.COLORMAP_JET)
    
    # 将热力图与原始图像混合
    alpha = 0.5
    # 确保原始图像是BGR格式用于OpenCV操作
    if len(original_image.shape) == 3 and original_image.shape[2] == 3:
        original_bgr = cv2.cvtColor(original_image, cv2.COLOR_RGB2BGR)
    else:
        original_bgr = cv2.cvtColor(original_image, cv2.COLOR_GRAY2BGR)
    
    overlayed = cv2.addWeighted(original_bgr, 1-alpha, heatmap, alpha, 0)
    
    # 二值化结果用于分割
    _, binary_mask = cv2.threshold(res_vis, 127, 255, cv2.THRESH_BINARY)
    segmented = cv2.bitwise_and(original_bgr, original_bgr, mask=binary_mask)
    
    # 转回RGB格式用于显示
    overlayed_rgb = cv2.cvtColor(overlayed, cv2.COLOR_BGR2RGB)
    segmented_rgb = cv2.cvtColor(segmented, cv2.COLOR_BGR2RGB)
    
    return original_image, res_vis, heatmap, overlayed_rgb, segmented_rgb, f"推理时间: {inference_time:.4f}秒"

def run_demo(input_image):
    """Gradio界面的主函数"""
    if input_image is None:
        return [None] * 5 + ["请上传图片"]
    
    # 处理图像
    original, saliency_map, heatmap, overlayed, segmented, time_info = process_image(input_image, model)
    
    return original, saliency_map, heatmap, overlayed, segmented, time_info

# 加载模型
print("正在加载模型...")
model = load_model()

# 创建Gradio界面
with gr.Blocks(title="显著性目标检测Demo") as demo:
    gr.Markdown("# 显著性目标检测Demo")
    gr.Markdown("上传一张图片，系统将自动检测显著性区域")
    
    with gr.Row():
        with gr.Column():
            input_image = gr.Image(label="输入图像", type="numpy")
            submit_btn = gr.Button("开始检测")
        
        with gr.Column():
            original_output = gr.Image(label="原始图像")
            saliency_output = gr.Image(label="显著性图")
            heatmap_output = gr.Image(label="热力图")
            overlayed_output = gr.Image(label="叠加结果")
            segmented_output = gr.Image(label="分割结果")
            time_info = gr.Textbox(label="处理信息")
    
    submit_btn.click(
        fn=run_demo,
        inputs=input_image,
        outputs=[original_output, saliency_output, heatmap_output, overlayed_output, segmented_output, time_info]
    )
    
    gr.Markdown("## 使用说明")
    gr.Markdown("1. 点击'输入图像'区域上传一张图片")
    gr.Markdown("2. 点击'开始检测'按钮进行显著性目标检测")
    gr.Markdown("3. 系统将显示原始图像、显著性图、热力图、叠加结果和分割结果")

# 启动Gradio应用
if __name__ == "__main__":
    demo.launch(share=True)