update

ultralytics/engine/exporter.py

@@ -268,6 +268,20 @@ class Exporter:
             elif isinstance(m, C2f) and not is_tf_format:
                 # EdgeTPU does not support FlexSplitV while split provides cleaner ONNX graph
                 m.forward = m.forward_split
+            if isinstance(m, Detect) and mct:
+                from ultralytics.utils.tal import make_anchors
+
+                anchors, strides = (
+                    x.transpose(0, 1)
+                    for x in make_anchors(
+                        torch.cat([s / m.stride.unsqueeze(-1) for s in self.imgsz], dim=1),
+                        m.stride,
+                        0.5,
+                    )
+                )
+                m.anchors = anchors
+                m.strides = strides
+
             if isinstance(m, C2f) and mct:
                 m.forward = m.forward_fx
 
@@ -1295,9 +1309,9 @@ class Exporter:
         model = ct.models.MLModel(pipeline.spec, weights_dir=weights_dir)
         model.input_description["image"] = "Input image"
         model.input_description["iouThreshold"] = f"(optional) IoU threshold override (default: {nms.iouThreshold})"
-        model.input_description["confidenceThreshold"] = (
-            f"(optional) Confidence threshold override (default: {nms.confidenceThreshold})"
-        )
+        model.input_description[
+            "confidenceThreshold"
+        ] = f"(optional) Confidence threshold override (default: {nms.confidenceThreshold})"
         model.output_description["confidence"] = 'Boxes × Class confidence (see user-defined metadata "classes")'
         model.output_description["coordinates"] = "Boxes × [x, y, width, height] (relative to image size)"
         LOGGER.info(f"{prefix} pipeline success")

ultralytics/nn/modules/head.py

@@ -89,13 +89,7 @@ class Detect(nn.Module):
         # Inference path
         shape = x[0].shape  # BCHW
         x_cat = torch.cat([xi.view(shape[0], self.no, -1) for xi in x], 2)
-        if self.export and self.format == "mct":
-            self.anchors, self.strides = (
-                x.transpose(0, 1)
-                for x in make_anchors(getattr(self, "feats_size", torch.Tensor([80, 40, 20])), self.stride, 0.5)
-            )
-            self.anchors, self.strides = self.anchors.to(x[0].device), self.strides.to(x[0].device)
-        elif self.dynamic or self.shape != shape:
+        if self.format != "mct" and (self.dynamic or self.shape != shape):
             self.anchors, self.strides = (x.transpose(0, 1) for x in make_anchors(x, self.stride, 0.5))
             self.shape = shape
 

ultralytics/utils/tal.py

@@ -306,7 +306,7 @@ def make_anchors(feats, strides, grid_cell_offset=0.5):
     assert feats is not None
     dtype, device = feats[0].dtype, feats[0].device
     for i, stride in enumerate(strides):
-        h, w = feats[i].shape[2:] if isinstance(feats, list) else (int(feats[i]), int(feats[i]))
+        h, w = feats[i].shape[2:] if isinstance(feats, list) else (int(feats[i][0]), int(feats[i][1]))
         sx = torch.arange(end=w, device=device, dtype=dtype) + grid_cell_offset  # shift x
         sy = torch.arange(end=h, device=device, dtype=dtype) + grid_cell_offset  # shift y
         sy, sx = torch.meshgrid(sy, sx, indexing="ij") if TORCH_1_10 else torch.meshgrid(sy, sx)