Fix TFLite error and OpenVINO int8 error at imgsz=32 (#18898)

Signed-off-by: Glenn Jocher <glenn.jocher@ultralytics.com>
Co-authored-by: Glenn Jocher <glenn.jocher@ultralytics.com>
Co-authored-by: UltralyticsAssistant <web@ultralytics.com>

docs/en/guides/azureml-quickstart.md

@@ -47,7 +47,7 @@ Start your compute and open a Terminal:
 ### Create virtualenv
 
 Create your conda virtualenv with your favorite python version and install pip in it:
-Python 3.13.1 is having some issues with some dependecies in AzureML.
+Python 3.13.1 is having some issues with some dependencies in AzureML.
 
 ```bash
 conda create --name yolo11env -y python=3.12
@@ -90,7 +90,7 @@ Open the compute Terminal.
   <img width="480" src="https://github.com/ultralytics/docs/releases/download/0/open-terminal.avif" alt="Open Terminal">
 </p>
 
-From your compute terminal, you need to create a new ipykernel (with a specific python version - because Python 3.13.1 is having some issues with some dependecies in AzureML) that will be used by your notebook to manage your dependencies:
+From your compute terminal, you need to create a new ipykernel (with a specific python version - because Python 3.13.1 is having some issues with some dependencies in AzureML) that will be used by your notebook to manage your dependencies:
 
 ```bash
 conda create --name yolo11env -y python=3.12

docs/mkdocs_github_authors.yaml

@@ -61,6 +61,9 @@
 48149018+zhixuwei@users.noreply.github.com:
   avatar: https://avatars.githubusercontent.com/u/48149018?v=4
   username: zhixuwei
+49172033+Lucashygi@users.noreply.github.com:
+  avatar: https://avatars.githubusercontent.com/u/49172033?v=4
+  username: Lucashygi
 49699333+dependabot[bot]@users.noreply.github.com:
   avatar: https://avatars.githubusercontent.com/u/27347476?v=4
   username: dependabot

tests/test_exports.py

@@ -81,7 +81,7 @@ def test_export_openvino_matrix(task, dynamic, int8, half, batch, nms):
         for task, dynamic, int8, half, batch, simplify, nms in product(
             TASKS, [True, False], [False], [False], [1, 2], [True, False], [True, False]
         )
-        if not ((int8 and half) or (task == "classify" and nms))
+        if not ((int8 and half) or (task == "classify" and nms) or (task == "obb" and nms and not TORCH_1_13))
     ],
 )
 def test_export_onnx_matrix(task, dynamic, int8, half, batch, simplify, nms):

ultralytics/engine/exporter.py

@@ -1559,6 +1559,7 @@ class NMSModel(torch.nn.Module):
         extra_shape = pred.shape[-1] - (4 + self.model.nc)  # extras from Segment, OBB, Pose
         boxes, scores, extras = pred.split([4, self.model.nc, extra_shape], dim=2)
         scores, classes = scores.max(dim=-1)
+        self.args.max_det = min(pred.shape[1], self.args.max_det)  # in case num_anchors < max_det
         # (N, max_det, 4 coords + 1 class score + 1 class label + extra_shape).
         out = torch.zeros(
             boxes.shape[0],
@@ -1596,14 +1597,25 @@ class NMSModel(torch.nn.Module):
                 offbox = nmsbox[:, :end] + cls_offset * multiplier
                 nmsbox = torch.cat((offbox, nmsbox[:, end:]), dim=-1)
             nms_fn = (
-                partial(nms_rotated, use_triu=not (self.is_tf or (self.args.opset or 14) < 14)) if self.obb else nms
+                partial(
+                    nms_rotated,
+                    use_triu=not (
+                        self.is_tf
+                        or (self.args.opset or 14) < 14
+                        or (self.args.format == "openvino" and self.args.int8)  # OpenVINO int8 error with triu
+                    ),
+                )
+                if self.obb
+                else nms
             )
             keep = nms_fn(
                 torch.cat([nmsbox, extra], dim=-1) if self.obb else nmsbox,
                 score,
                 self.args.iou,
             )[: self.args.max_det]
-            dets = torch.cat([box[keep], score[keep].view(-1, 1), cls[keep].view(-1, 1), extra[keep]], dim=-1)
+            dets = torch.cat(
+                [box[keep], score[keep].view(-1, 1), cls[keep].view(-1, 1).to(out.dtype), extra[keep]], dim=-1
+            )
             # Zero-pad to max_det size to avoid reshape error
             pad = (0, 0, 0, self.args.max_det - dets.shape[0])
             out[i] = torch.nn.functional.pad(dets, pad)