import cv2 import numpy as np from tensorflow.keras.models import load_model from tensorflow.keras.preprocessing.image import img_to_array import gradio as gr import json # ========================================== # 1. 加载模型与标签 # ========================================== emotion_model = load_model('ferplus_cnn_best.h5') emotion_labels = ['anger', 'contempt', 'disgust', 'fear', 'happiness', 'neutral', 'sadness', 'surprise'] # 载入复合情绪规则 try: with open('compound_rules.json', 'r', encoding='utf-8') as f: compound_rules = json.load(f)['compound_emotions'] except FileNotFoundError: print("[Error] compound_rules.json not found! Please upload it.") compound_rules = [] # 加载人脸检测器 net = cv2.dnn.readNetFromCaffe("deploy.prototxt", "res10_300x300_ssd_iter_140000.caffemodel") # ========================================== # 2. 复合情绪逻辑 # ========================================== def get_compound_emotion(predictions): for rule in compound_rules: meets = True for cond in rule['conditions']: idx = emotion_labels.index(cond['emotion']) if predictions[idx] < cond['min']: meets = False break if meets: conf = np.mean([predictions[emotion_labels.index(c['emotion'])] for c in rule['conditions']]) return rule['name'], float(conf) return None, None # ========================================== # 3. 核心检测逻辑 (适配网页) # ========================================== def predict_emotion(input_img): if input_img is None: return None, "Please upload or take a photo first." # Gradio RGB 转 OpenCV BGR frame = cv2.cvtColor(input_img, cv2.COLOR_RGB2BGR) h, w = frame.shape[:2] blob = cv2.dnn.blobFromImage(frame, 1.0, (300, 300), (104.0, 177.0, 123.0)) net.setInput(blob) detections = net.forward() confidences = {label: 0.0 for label in emotion_labels} faces_found = False for i in range(detections.shape[2]): confidence = detections[0, 0, i, 2] if confidence > 0.5: faces_found = True box = detections[0, 0, i, 3:7] * np.array([w, h, w, h]) (x, y, x2, y2) = box.astype("int") x, y = max(0, x), max(0, y) x2, y2 = min(w, x2), min(h, y2) gray_frame = cv2.cvtColor(frame[y:y2, x:x2], cv2.COLOR_BGR2GRAY) if gray_frame.size == 0: continue face_roi = cv2.resize(gray_frame, (112, 112)) face_roi = face_roi.astype('float32') / 255.0 face_roi = img_to_array(face_roi) face_roi = np.expand_dims(face_roi, axis=0) # 模型预测基础情绪 preds = emotion_model.predict(face_roi, verbose=0)[0] # 更新右侧统计图表的数据 (始终显示8个基础情绪的原始分布) confidences = {emotion_labels[i]: float(preds[i]) for i in range(len(emotion_labels))} # 复合情绪判断 compound_name, compound_conf = get_compound_emotion(preds) if compound_name: label_text = f"{compound_name} ({compound_conf * 100:.1f}%)" color = (0, 255, 0) if compound_conf > 0.7 else (0, 255, 255) if compound_conf > 0.4 else (0, 0, 255) else: emotion_index = np.argmax(preds) emotion_label = emotion_labels[emotion_index] conf = preds[emotion_index] label_text = f"{emotion_label} ({conf * 100:.1f}%)" color = (0, 255, 0) if conf > 0.7 else (0, 255, 255) if conf > 0.4 else (0, 0, 255) # 画框和文字 cv2.rectangle(frame, (x, y), (x2, y2), color, 3) cv2.putText(frame, label_text, (x, y-10), cv2.FONT_HERSHEY_SIMPLEX, 0.8, color, 2) if not faces_found: return cv2.cvtColor(frame, cv2.COLOR_BGR2RGB), "No face detected" # 返回处理后的图片 和 用于生成进度条的字典 return cv2.cvtColor(frame, cv2.COLOR_BGR2RGB), confidences # ========================================== # 4. Gradio 界面设计 # ========================================== with gr.Blocks(theme=gr.themes.Default()) as demo: gr.Markdown("# Facial Emotion Recognition") gr.Markdown("Identify complex human emotions (e.g., Frustrated, Embarrassed) from images or webcam using customized JSON rules.") with gr.Row(): with gr.Column(): input_image = gr.Image(label="Input Source", sources=["upload", "webcam"], type="numpy") run_btn = gr.Button("Analyze Emotion", variant="primary") with gr.Column(): output_image = gr.Image(label="Detection Result") output_labels = gr.Label(label="Base Emotion Probabilities", num_top_classes=4) run_btn.click(fn=predict_emotion, inputs=input_image, outputs=[output_image, output_labels]) if __name__ == "__main__": demo.launch()