Update app.py

app.py

@@ -8,7 +8,6 @@ import requests
 import os
 from typing import Dict, List, Tuple
 import asyncio
-import aiohttp
 
 # Initialize body estimation model
 body_estimation = Body('model/body_pose_model.pth')
@@ -67,9 +66,9 @@ def pil2cv(image):
         new_image = cv2.cvtColor(new_image, cv2.COLOR_RGBA2BGRA)
     return new_image
 
-async def generate_pose_from_llm(prompt: str) -> Dict:
+def generate_pose_from_llm_sync(prompt: str) -> Dict:
     """
-    LLM을 사용하여 텍스트 프롬프트로부터 포즈 데이터를 생성
+    LLM을 사용하여 텍스트 프롬프트로부터 포즈 데이터를 생성 (동기 버전)
     """
     system_prompt = """You are an expert in human pose generation. Given a description, generate precise OpenPose keypoint coordinates.
     
@@ -109,23 +108,21 @@ async def generate_pose_from_llm(prompt: str) -> Dict:
     }
     
     try:
-        async with aiohttp.ClientSession() as session:
-            async with session.post(FIREWORKS_API_URL, headers=headers, json=payload) as response:
-                if response.status == 200:
-                    data = await response.json()
-                    content = data['choices'][0]['message']['content']
-                    
-                    # Extract JSON from response
-                    import re
-                    json_match = re.search(r'\{.*\}', content, re.DOTALL)
-                    if json_match:
-                        pose_data = js.loads(json_match.group())
-                        return pose_data
-                    else:
-                        # Fallback to template
-                        return generate_template_pose(prompt)
-                else:
-                    return generate_template_pose(prompt)
+        response = requests.post(FIREWORKS_API_URL, headers=headers, json=payload, timeout=30)
+        if response.status_code == 200:
+            data = response.json()
+            content = data['choices'][0]['message']['content']
+            
+            # Extract JSON from response
+            import re
+            json_match = re.search(r'\{.*\}', content, re.DOTALL)
+            if json_match:
+                pose_data = js.loads(json_match.group())
+                return pose_data
+            else:
+                return generate_template_pose(prompt)
+        else:
+            return generate_template_pose(prompt)
     except Exception as e:
         print(f"LLM Error: {e}")
         return generate_template_pose(prompt)
@@ -149,18 +146,31 @@ def generate_template_pose(prompt: str) -> Dict:
     for i in range(18):
         if i == 0:  # Nose
             candidate.append([256, 100, 1.0])
-        elif part_name := next((k for k, v in BODY_PARTS.items() if v == i), None):
-            if part_name in template["keypoints"]:
+        elif i == 14:  # REye
+            candidate.append([246, 90, 1.0])
+        elif i == 15:  # LEye
+            candidate.append([266, 90, 1.0])
+        elif i == 16:  # REar
+            candidate.append([236, 95, 1.0])
+        elif i == 17:  # LEar
+            candidate.append([276, 95, 1.0])
+        else:
+            # Find part name for this index
+            part_name = None
+            for name, idx in BODY_PARTS.items():
+                if idx == i:
+                    part_name = name
+                    break
+            
+            if part_name and part_name in template["keypoints"]:
                 x, y = template["keypoints"][part_name]
                 candidate.append([x, y, 1.0])
             else:
-                # Estimate position based on nearby keypoints
                 candidate.append([256, 256, 0.0])
-        else:
-            candidate.append([0, 0, 0.0])
     
     # Create subset (connection information)
-    subset = [[i for i in range(18) if candidate[i][2] > 0] + [18.0, 18]]
+    valid_indices = [i for i in range(18) if candidate[i][2] > 0]
+    subset = [valid_indices + [float(len(valid_indices)), len(valid_indices)]]
     
     return {"candidate": candidate, "subset": subset}
 
@@ -190,7 +200,7 @@ def refine_pose_with_llm(current_pose: Dict, refinement_prompt: str) -> Dict:
     }
     
     try:
-        response = requests.post(FIREWORKS_API_URL, headers=headers, json=payload)
+        response = requests.post(FIREWORKS_API_URL, headers=headers, json=payload, timeout=30)
         if response.status_code == 200:
             data = response.json()
             content = data['choices'][0]['message']['content']
@@ -205,8 +215,11 @@ def refine_pose_with_llm(current_pose: Dict, refinement_prompt: str) -> Dict:
     return current_pose
 
 # FastAPI setup
-with open("static/poseEditor.js", "r") as f:
-    file_contents = f.read()
+try:
+    with open("static/poseEditor.js", "r") as f:
+        file_contents = f.read()
+except:
+    file_contents = "console.log('PoseEditor.js not found');"
 
 app = FastAPI()
 
@@ -239,12 +252,23 @@ async def some_fastapi_middleware(request: Request, call_next):
     return response
 
 def candidate_to_json_string(arr):
-    a = [f'[{x:.2f}, {y:.2f}]' for x, y, *_ in arr]
-    return '[' + ', '.join(a) + ']'
+    if isinstance(arr, list):
+        a = []
+        for item in arr:
+            if len(item) >= 2:
+                x, y = item[0], item[1]
+                a.append(f'[{float(x):.2f}, {float(y):.2f}]')
+        return '[' + ', '.join(a) + ']'
+    return '[]'
 
 def subset_to_json_string(arr):
-    arr_str = ','.join(['[' + ','.join([f'{num:.2f}' for num in row]) + ']' for row in arr])
-    return '[' + arr_str + ']'
+    if isinstance(arr, np.ndarray):
+        arr_str = ','.join(['[' + ','.join([f'{num:.2f}' for num in row]) + ']' for row in arr])
+        return '[' + arr_str + ']'
+    elif isinstance(arr, list):
+        arr_str = ','.join(['[' + ','.join([f'{float(num):.2f}' for num in row]) + ']' for row in arr])
+        return '[' + arr_str + ']'
+    return '[]'
 
 def estimate_body(source):
     if source == None:
@@ -269,25 +293,18 @@ def image_changed(image):
         jsonText = "{ \"candidate\": " + candidate_to_json_string(candidate) + ", \"subset\": " + subset_to_json_string(subset) + " }"
         return f"""{image.width}px x {image.height}px, {subset.shape[0]} individual(s)""", jsonText
 
-async def generate_pose_from_text(prompt: str, use_llm: bool = True):
+def generate_pose_from_text(prompt: str, use_llm: bool = True):
     """
-    텍스트 프롬프트로부터 포즈 생성
+    텍스트 프롬프트로부터 포즈 생성 (동기 버전)
     """
     if use_llm and FIREWORKS_API_KEY != "YOUR_API_KEY_HERE":
-        pose_data = await generate_pose_from_llm(prompt)
+        pose_data = generate_pose_from_llm_sync(prompt)
     else:
         pose_data = generate_template_pose(prompt)
     
     # Format for the pose editor
-    if isinstance(pose_data['candidate'], list):
-        candidate_str = candidate_to_json_string(pose_data['candidate'])
-    else:
-        candidate_str = js.dumps(pose_data['candidate'])
-    
-    if isinstance(pose_data['subset'], list):
-        subset_str = subset_to_json_string(pose_data['subset'])
-    else:
-        subset_str = js.dumps(pose_data['subset'])
+    candidate_str = candidate_to_json_string(pose_data['candidate'])
+    subset_str = subset_to_json_string(pose_data['subset'])
     
     return "{ \"candidate\": " + candidate_str + ", \"subset\": " + subset_str + " }"
 
@@ -298,9 +315,9 @@ html_text = f"""
 
 # Gradio interface
 with gr.Blocks(css="""
-    button { min-width: 80px; }
-    .prompt-box { border: 2px solid #667eea; border-radius: 8px; padding: 10px; }
-    .llm-status { color: #667eea; font-weight: bold; }
+    button {{ min-width: 80px; }}
+    .prompt-box {{ border: 2px solid #667eea; border-radius: 8px; padding: 10px; }}
+    .llm-status {{ color: #667eea; font-weight: bold; }}
     """) as demo:
     
     gr.Markdown("""
@@ -434,19 +451,16 @@ with gr.Blocks(css="""
     )
     
     # LLM generation events
-    async def handle_generate(prompt, use_llm):
+    def handle_generate(prompt, use_llm):
         if not prompt:
             return None, "⚠️ Please enter a pose description"
         
         try:
             status = "🔄 Generating pose with AI..." if use_llm else "🔄 Using template..."
-            yield None, status
-            
-            pose_json = await generate_pose_from_text(prompt, use_llm)
-            yield pose_json, "✅ Pose generated successfully!"
-            
+            pose_json = generate_pose_from_text(prompt, use_llm)
+            return pose_json, "✅ Pose generated successfully!"
         except Exception as e:
-            yield None, f"❌ Error: {str(e)}"
+            return None, f"❌ Error: {str(e)}"
     
     generate_btn.click(
         fn=handle_generate,
@@ -467,7 +481,7 @@ with gr.Blocks(css="""
         outputs=[refinement_prompt]
     )
     
-    async def handle_refine(current_json, refinement):
+    def handle_refine(current_json, refinement):
         if not current_json or not refinement:
             return None, "⚠️ Need current pose and refinement instructions"
         
@@ -530,4 +544,10 @@ with gr.Blocks(css="""
     
     demo.load(fn=check_api_status, outputs=[llm_status])
 
-gr.mount_gradio_app(app, demo, path="/")
+# Mount Gradio app to FastAPI
+gr.mount_gradio_app(app, demo, path="/")
+
+# For running the app
+if __name__ == "__main__":
+    import uvicorn
+    uvicorn.run(app, host="0.0.0.0", port=7860)