Update garb cls to render on colors and infer 3 models together

app.py

@@ -423,49 +423,70 @@ async def classify_garbage(file: UploadFile = File(...)):
     img_id = _uid()
     img_path = f"{UPLOAD_DIR}/{img_id}_{file.filename}"
     out_path = f"{OUTPUT_DIR}/{img_id}_classified.jpg"
-    # Load file
+    # Save uploaded file
     with open(img_path, "wb") as f:
         shutil.copyfileobj(file.file, f)
-    # Read image
+    # Read file
     print(f"[Classification] Received image: {img_path}")
     image = cv2.imread(img_path)
     rgb = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
     pil = Image.fromarray(rgb)
-    # DETR for garbage detection boxes
+    # ─── Detection from 3 models ─────────────────────────────
     detections = []
-    inp = processor_detr(images=pil, return_tensors="pt")
+    # YOLOv11 (self-trained)
+    for r in model_self(image):
+        detections += [b.xyxy[0].tolist() for b in r.boxes]
+    # YOLOv5
+    r = model_yolo5(image)
+    if hasattr(r, 'pred') and len(r.pred) > 0:
+        detections += [p[:4].tolist() for p in r.pred[0]]
+    # DETR
     with torch.no_grad():
-        out = model_detr(**inp)
+        out = model_detr(**processor_detr(images=pil, return_tensors="pt"))
     results = processor_detr.post_process_object_detection(
         outputs=out,
         target_sizes=torch.tensor([pil.size[::-1]]),
         threshold=0.5
     )[0]
-    # Bbox return
-    boxes = results["boxes"]
-    print(f"[Classification] {len(boxes)} garbage objects detected by DETR.")
-    # Mapping in between
-    for i, box in enumerate(boxes):
-        x1, y1, x2, y2 = map(int, box.tolist())
+    detections += [b.tolist() for b in results["boxes"]]
+    print(f"[Classification] Total detections from 3 models: {len(detections)}")
+    # ─── Classification & Rendering ─────────────────────────
+    for box in detections:
+        x1, y1, x2, y2 = map(int, box)
+        x1, x2 = max(0, min(x1, 639)), max(0, min(x2, 639))
+        y1, y2 = max(0, min(y1, 639)), max(0, min(y2, 639))
+        # Stack all crops
         crop = image[y1:y2, x1:x2]
         if crop.shape[0] < 10 or crop.shape[1] < 10:
-            continue  # skip tiny crops
-        # Convert crop to RGB and classify
+            continue
+        # Image processing
         pil_crop = Image.fromarray(cv2.cvtColor(crop, cv2.COLOR_BGR2RGB))
-        pred = model_garbage_cls(pil_crop, verbose=False)[0]
+        with torch.no_grad():
+            pred = model_garbage_cls(pil_crop, verbose=False)[0]
         class_id = int(pred.probs.top1)
         class_name = model_garbage_cls.names[class_id]
-        conf = pred.probs.top1conf
-        # Labelling on output image
+        conf = float(pred.probs.top1conf)
+        # Label format
         label = f"{class_name} ({conf:.2f})"
-        cv2.rectangle(image, (x1, y1), (x2, y2), (0, 165, 255), 2)
-        cv2.putText(image, label, (x1, y1 - 5), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 165, 255), 2)
-    # Write image on render
+        # Dynamic color coding
+        if conf < 0.4:
+            color = (0, 0, 255)       # Red
+        elif conf < 0.6:
+            color = (0, 255, 0)       # Green
+        elif conf < 0.8:
+            color = (255, 255, 0)     # Sky Blue
+        else:
+            color = (255, 0, 255)     # Purple
+        # Labelling
+        cv2.rectangle(image, (x1, y1), (x2, y2), color, 2)
+        cv2.putText(image, label, (x1, y1 - 6), cv2.FONT_HERSHEY_SIMPLEX, 0.6, color, 2)
+    # Save result
     cv2.imwrite(out_path, image)
     print(f"[Classification] Output saved: {out_path}")
     return FileResponse(out_path, media_type="image/jpeg")
 
 
+
 # ── Core pipeline (runs in background thread) ───────────────────────────
 def _pipeline(uid,img_path):
     print(f"▶️ [{uid}] processing")