Upload folder using huggingface_hub

miner.py

@@ -72,7 +72,7 @@ class Miner:
         self.input_width = self._safe_dim(self.input_shape[3], default=1280)
 
         self.conf_thres = 0.25
-        self.iou_thres = 0.5
+        self.iou_thres = 0.45
         self.max_det = 300
 
         print(f"✅ ONNX model loaded from: {model_path}")
@@ -265,7 +265,7 @@ class Miner:
         """
         orig_h, orig_w = image.shape[:2]
 
-        # image = self._enhance_image(image)
+        image = self._enhance_image(image)
 
         img, ratio, pad = self._letterbox(
             image, (self.input_width, self.input_height)
@@ -369,13 +369,36 @@ class Miner:
         boxes /= ratio
         boxes = self._clip_boxes(boxes, (orig_w, orig_h))
 
-        # Optional dedup if exported model still emits overlaps
+        # Dedup: standard NMS + containment check
         if apply_optional_dedup and len(boxes) > 1:
             keep_idx = self._nms(boxes, scores, self.iou_thres)
             boxes = boxes[keep_idx]
             scores = scores[keep_idx]
             cls_ids = cls_ids[keep_idx]
 
+        # Remove boxes where a smaller box is mostly contained in a larger one
+        if len(boxes) > 1:
+            keep = np.ones(len(boxes), dtype=bool)
+            areas = (boxes[:, 2] - boxes[:, 0]) * (boxes[:, 3] - boxes[:, 1])
+            for i in range(len(boxes)):
+                if not keep[i]:
+                    continue
+                for j in range(i + 1, len(boxes)):
+                    if not keep[j]:
+                        continue
+                    xx1 = np.maximum(boxes[i, 0], boxes[j, 0])
+                    yy1 = np.maximum(boxes[i, 1], boxes[j, 1])
+                    xx2 = np.minimum(boxes[i, 2], boxes[j, 2])
+                    yy2 = np.minimum(boxes[i, 3], boxes[j, 3])
+                    inter = max(0.0, xx2 - xx1) * max(0.0, yy2 - yy1)
+                    smaller_area = min(areas[i], areas[j])
+                    if smaller_area > 0 and inter / smaller_area > 0.6:
+                        victim = j if scores[i] >= scores[j] else i
+                        keep[victim] = False
+            boxes = boxes[keep]
+            scores = scores[keep]
+            cls_ids = cls_ids[keep]
+
         results: list[BoundingBox] = []
         for box, conf, cls_id in zip(boxes, scores, cls_ids):
             x1, y1, x2, y2 = box.tolist()

weights.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:26b9faebedf842fd6367928ea2a07cff7eb25910e2f6871110bd500d0c544f3e
-size 19405465
+oid sha256:be6e4b7e468f655482d0f73509954721601eacb68d376d8191a14bdb7f3d3105
+size 19404973