Update app.py

app.py

@@ -2,24 +2,88 @@ import gradio as gr
 import json
 import base64
 import os
+import requests
+from huggingface_hub import HfApi, HfFolder
+
+HF_TOKEN = os.environ.get("HF_CV_ROBOT_TOKEN")
+MODEL = "Qwen/Qwen2.5-VL-7B-Instruct"
+REPO_ID = "your-username/tmp-vlm-images"  # Replace with your HF repo
+
+if not HF_TOKEN:
+    print("ERROR: HF_CV_ROBOT_TOKEN environment variable not set.")
+
+api = HfApi()
+HfFolder.save_token(HF_TOKEN)
+
+def upload_to_hf(filepath: str):
+    filename = os.path.basename(filepath)
+    # Upload to HF repo
+    api.upload_file(
+        path_or_fileobj=filepath,
+        path_in_repo=filename,
+        repo_id=REPO_ID,
+        repo_type="dataset",
+        token=HF_TOKEN,
+        overwrite=True
+    )
+    # Construct raw URL
+    url = f"https://huggingface.co/datasets/{REPO_ID}/resolve/main/{filename}"
+    return url
 
 def process(payload: dict):
     try:
+        if not HF_TOKEN:
+            return {"error": "Hugging Face token missing."}
+
+        robot_id = payload.get("robot_id", "unknown")
         image_b64 = payload["image_b64"]
 
-        # 直接寫到 space sandbox
+        # 1️⃣ Save the image locally
         tmp_path = "/tmp/tmp.jpg"
-
         with open(tmp_path, "wb") as f:
             f.write(base64.b64decode(image_b64))
-
-        # 再讀取確認寫入成功
         file_size = os.path.getsize(tmp_path)
 
+        # 2️⃣ Upload to HF to get URL
+        image_url = upload_to_hf(tmp_path)
+
+        # 3️⃣ Call VLM with image URL
+        data = {
+            "model": MODEL,
+            "messages": [
+                {
+                    "role": "user",
+                    "content": [
+                        {"type": "text", "text": "Describe this image in detail."},
+                        {"type": "image_url", "image_url": image_url}
+                    ]
+                }
+            ]
+        }
+
+        resp = requests.post(
+            "https://router.huggingface.co/v1/chat/completions",
+            headers={"Authorization": f"Bearer {HF_TOKEN}"},
+            json=data,
+            timeout=60
+        )
+
+        if resp.status_code != 200:
+            vlm_text = f"HF VLM error: {resp.status_code}, {resp.text}"
+        else:
+            try:
+                vlm_text = resp.json()["choices"][0]["message"]["content"][0]["text"]
+            except Exception as e:
+                vlm_text = f"Failed to parse VLM response: {e}, Response={resp.text}"
+
+        # 4️⃣ Return combined info
         return {
             "saved": True,
             "file_path": tmp_path,
-            "file_size_bytes": file_size
+            "file_size_bytes": file_size,
+            "robot_id": robot_id,
+            "vlm_description": vlm_text,
+            "image_url": image_url
         }
 
     except Exception as e:
@@ -28,7 +92,7 @@ def process(payload: dict):
 demo = gr.Interface(
     fn=process,
     inputs=gr.JSON(label="Input Payload (Dict format)"),
-    outputs=gr.JSON(label="Reply"),
+    outputs=gr.JSON(label="Reply to Jetson"),
     api_name="predict"
 )