app.py

import gradio as gr
import mediapipe as mp
import cv2
import numpy as np
from PIL import Image
import torch
from transformers import AutoProcessor, AutoTokenizer, AutoModelForCausalLM
from posture_utils import analyze_posture_by_keypoints

model_id = "Qwen/Qwen-VL-Chat"
tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
processor = AutoProcessor.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(model_id, trust_remote_code=True, device_map="auto").eval()

mp_pose = mp.solutions.pose

def process(image: Image):
    np_image = np.array(image)
    with mp_pose.Pose(static_image_mode=True) as pose:
        results = pose.process(cv2.cvtColor(np_image, cv2.COLOR_RGB2BGR))

        if not results.pose_landmarks:
            return "❗ 无法检测到人体，请上传包含上半身的清晰坐姿照片。"

        posture_analysis = analyze_posture_by_keypoints(results.pose_landmarks)
        prompt = f"<|im_start|>user\n请根据以下坐姿描述生成中英文提醒：\n{posture_analysis}\n<|im_end|>\n<|im_start|>assistant\n"
        inputs = processor(images=image, text=prompt, return_tensors="pt").to(model.device)
        output = model.generate(**inputs, max_new_tokens=512)
        result = tokenizer.decode(output[0], skip_special_tokens=True)
        return result.split("<|im_start|>assistant\n")[-1].strip()

demo = gr.Interface(
    fn=process,
    inputs=gr.Image(type="pil", label="上传你的坐姿照片"),
    outputs=gr.Textbox(label="中英文坐姿分析结果"),
    title="🪑 坐姿监测融合助手",
    description="上传坐姿图像，先通过 Mediapipe 判断是否驼背、低头、含胸，再交由 Qwen-VL 生成提醒语。",
    theme="soft",
    allow_flagging="never"
)

if __name__ == "__main__":
    demo.launch()