Update app.py

app.py

@@ -1,59 +1,68 @@
 import gradio as gr
-import base64
 import json
+import base64
+from PIL import Image
+import io
 import requests
 import os
 
-HF_ROUTER_API = "https://router.huggingface.co/hf-inference"
 HF_TOKEN = os.getenv("HF_CV_ROBOT_TOKEN")
-MODEL_NAME = "Qwen/Qwen3-VL-32B-Instruct"
-
-def call_vlm_api(payload: dict):
-    """
-    Call Hugging Face Router Inference API with Base64 image.
-    """
-    headers = {"Authorization": f"Bearer {HF_TOKEN}"}
-    
-    data = {
-        "model": MODEL_NAME,
-        "inputs": [
-            {
-                "image": {"b64": payload["image_b64"]},
-                "text": "Describe the image in detail."
-            }
-        ]
-    }
-    
-    try:
-        resp = requests.post(HF_ROUTER_API, headers=headers, json=data, timeout=60)
-        if resp.status_code == 200:
-            # 取第一個 generated_text
-            return resp.json()[0].get("generated_text", "")
-        else:
-            return f"VLM API error: {resp.status_code}, {resp.text}"
-    except Exception as e:
-        return f"Exception: {str(e)}"
+MODEL = "Qwen/Qwen3-VL-32B-Instruct"
 
+# -------------------------------
+# 主處理函數
+# -------------------------------
 def process(payload: dict):
-    """
-    Process JSON payload from Jetson: Base64 image + robot_id
-    Return JSON with VLM analysis
-    """
     try:
-        vlm_text = call_vlm_api(payload)
-        reply = {
+        robot_id = payload.get("robot_id", "unknown")
+        image_b64 = payload["image_b64"]
+
+        # Base64 解碼成圖片，用 PIL 開啟
+        img_bytes = base64.b64decode(image_b64)
+        img = Image.open(io.BytesIO(img_bytes)).convert("RGB")
+
+        # Router API payload
+        headers = {"Authorization": f"Bearer {HF_TOKEN}"}
+        data = {
+            "model": MODEL,
+            "messages": [
+                {
+                    "role": "user",
+                    "content": [
+                        {"type": "text", "text": "Describe this image in detail."},
+                        {"type": "image_data", "image_data": {"b64": image_b64}}
+                    ]
+                }
+            ]
+        }
+
+        resp = requests.post(
+            "https://router.huggingface.co/v1/chat/completions",
+            headers=headers,
+            json=data,
+            timeout=60
+        )
+
+        if resp.status_code != 200:
+            return {"error": f"VLM API error: {resp.status_code}, {resp.text}"}
+
+        vlm_text = resp.json()["choices"][0]["message"]["content"][0]["text"]
+
+        return {
             "received": True,
-            "robot_id": payload.get("robot_id", "unknown"),
+            "robot_id": robot_id,
             "vllm_analysis": vlm_text
         }
-        return reply
+
     except Exception as e:
         return {"error": str(e)}
 
-# Gradio MCP server
+# -------------------------------
+# Gradio MCP Server
+# -------------------------------
 demo = gr.Interface(
     fn=process,
-    inputs=gr.JSON(label="Input Payload from Jetson"),
+    inputs=gr.JSON(label="Input Payload (Dict format)"),
     outputs=gr.JSON(label="Reply to Jetson"),
     api_name="predict"
 )