Replace inference-sdk with requests, pin Python 3.10

README.md

@@ -4,6 +4,7 @@ emoji: 🚦
 colorFrom: red
 colorTo: blue
 sdk: gradio
+python_version: "3.10"
 app_file: app.py
 pinned: false
 ---

requirements.txt

@@ -45,4 +45,3 @@ scipy
 packaging
 filelock==3.29.0
 gradio
-inference-sdk

solution.py

@@ -27,14 +27,38 @@ from transformers import pipeline as hf_pipeline
 from ultralytics import YOLO
 from paddleocr import PaddleOCR
 
-try:
-    from inference_sdk import InferenceHTTPClient
-    CLIENT = InferenceHTTPClient(
-        api_url="https://serverless.roboflow.com",
-        api_key="SEsiEStxDAHdOx2SCo3k"
-    )
-except ImportError:
-    CLIENT = None
+import requests as _requests
+import base64 as _base64
+
+ROBOFLOW_API_KEY = "SEsiEStxDAHdOx2SCo3k"
+
+def _roboflow_infer(image_input, model_id):
+    """Call the Roboflow serverless API using plain requests (no inference-sdk needed)."""
+    try:
+        if isinstance(image_input, str):
+            with open(image_input, "rb") as f:
+                img_bytes = f.read()
+        elif isinstance(image_input, np.ndarray):
+            _, buf = cv2.imencode(".jpg", image_input)
+            img_bytes = buf.tobytes()
+        else:
+            return {}
+        img_b64 = _base64.b64encode(img_bytes).decode("utf-8")
+        url = f"https://serverless.roboflow.com/{model_id}"
+        resp = _requests.post(
+            url,
+            params={"api_key": ROBOFLOW_API_KEY},
+            json={"image": [{"type": "base64", "value": img_b64}]},
+            timeout=30,
+        )
+        # Try v2 response format first, fall back to v1
+        data = resp.json()
+        if isinstance(data, list) and len(data) > 0:
+            return data[0]
+        return data
+    except Exception as e:
+        print(f"[Warning] Roboflow API error for {model_id}: {e}")
+        return {}
 
 # ── CONSTANTS ─────────────────────────────────────────────────────────────────
 COCO_PERSON = 0
@@ -329,17 +353,16 @@ class TrafficViolationDetector:
 
             # Detect Wrong Way using Roboflow API
             ww_boxes = []
-            if CLIENT is not None:
-                try:
-                    result = CLIENT.infer(img, model_id="wrong-way-driving-detection-gqdmg/1")
-                    for pred in result.get('predictions', []):
-                        if "wrong" in pred.get('class', '').lower():
-                            px, py, pw, ph = pred['x'], pred['y'], pred['width'], pred['height']
-                            wx1, wy1 = px - pw/2, py - ph/2
-                            wx2, wy2 = px + pw/2, py + ph/2
-                            ww_boxes.append([wx1, wy1, wx2, wy2])
-                except Exception as e:
-                    print("[Warning] Wrong-way API error:", e)
+            try:
+                result = _roboflow_infer(img, "wrong-way-driving-detection-gqdmg/1")
+                for pred in result.get('predictions', []):
+                    if "wrong" in pred.get('class', '').lower():
+                        px, py, pw, ph = pred['x'], pred['y'], pred['width'], pred['height']
+                        wx1, wy1 = px - pw/2, py - ph/2
+                        wx2, wy2 = px + pw/2, py + ph/2
+                        ww_boxes.append([wx1, wy1, wx2, wy2])
+            except Exception as e:
+                print("[Warning] Wrong-way API error:", e)
             
             def is_wrong_way(v_box):
                 for wb in ww_boxes:
@@ -376,17 +399,16 @@ class TrafficViolationDetector:
                 ww = is_wrong_way(car_box)
                 
                 sb_viols = 0
-                if CLIENT is not None:
-                    ccrop = img[max(0, y1):min(h_img, y2), max(0, x1):min(w_img, x2)]
-                    if ccrop.size > 0:
-                        try:
-                            res = CLIENT.infer(ccrop, model_id="seat-belt-detection-udcfg/5")
-                            for pred in res.get('predictions', []):
-                                cls_name = pred.get('class', '').lower()
-                                if "no" in cls_name and "seatbelt" in cls_name:
-                                    sb_viols += 1
-                        except Exception as e:
-                            print("[Warning] Seatbelt API error:", e)
+                ccrop = img[max(0, y1):min(h_img, y2), max(0, x1):min(w_img, x2)]
+                if ccrop.size > 0:
+                    try:
+                        res = _roboflow_infer(ccrop, "seat-belt-detection-udcfg/5")
+                        for pred in res.get('predictions', []):
+                            cls_name = pred.get('class', '').lower()
+                            if "no" in cls_name and "seatbelt" in cls_name:
+                                sb_viols += 1
+                    except Exception as e:
+                        print("[Warning] Seatbelt API error:", e)
                 
                 # Check for violation first, then do plate OCR if violation exists
                 if sb_viols > 0 or ww: