ultralytics 8.3.214__tar.gz → 8.3.215__tar.gz

{ultralytics-8.3.214 → ultralytics-8.3.215}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ultralytics
-Version: 8.3.214
+Version: 8.3.215
 Summary: Ultralytics YOLO 🚀 for SOTA object detection, multi-object tracking, instance segmentation, pose estimation and image classification.
 Author-email: Glenn Jocher <glenn.jocher@ultralytics.com>, Jing Qiu <jing.qiu@ultralytics.com>
 Maintainer-email: Ultralytics <hello@ultralytics.com>

{ultralytics-8.3.214 → ultralytics-8.3.215}/ultralytics/__init__.py

@@ -1,6 +1,6 @@
 # Ultralytics 🚀 AGPL-3.0 License - https://ultralytics.com/license
 
-__version__ = "8.3.214"
+__version__ = "8.3.215"
 
 import importlib
 import os

{ultralytics-8.3.214 → ultralytics-8.3.215}/ultralytics/engine/exporter.py

@@ -385,7 +385,7 @@ class Exporter:
             assert not tflite or not ARM64 or not LINUX, "TFLite export with NMS unsupported on ARM64 Linux"
             assert not is_tf_format or TORCH_1_13, "TensorFlow exports with NMS require torch>=1.13"
             assert not onnx or TORCH_1_13, "ONNX export with NMS requires torch>=1.13"
-            if getattr(model, "end2end", False):
+            if getattr(model, "end2end", False) or isinstance(model.model[-1], RTDETRDecoder):
                 LOGGER.warning("'nms=True' is not available for end2end models. Forcing 'nms=False'.")
                 self.args.nms = False
             self.args.conf = self.args.conf or 0.25  # set conf default value for nms export
@@ -1039,7 +1039,7 @@ class Exporter:
             attempt_download_asset(f"{onnx2tf_file}.zip", unzip=True, delete=True)
 
         # Export to ONNX
-        if "rtdetr" in self.model.model[-1]._get_name().lower():
+        if isinstance(self.model.model[-1], RTDETRDecoder):
             self.args.opset = self.args.opset or 19
             assert 16 <= self.args.opset <= 19, "RTDETR export requires opset>=16;<=19"
         self.args.simplify = True

{ultralytics-8.3.214 → ultralytics-8.3.215}/ultralytics/models/yolo/segment/predict.py

@@ -108,6 +108,7 @@ class SegmentationPredictor(DetectionPredictor):
             masks = ops.process_mask(proto, pred[:, 6:], pred[:, :4], img.shape[2:], upsample=True)  # HWC
             pred[:, :4] = ops.scale_boxes(img.shape[2:], pred[:, :4], orig_img.shape)
         if masks is not None:
-            keep = masks.sum((-2, -1)) > 0  # only keep predictions with masks
-            pred, masks = pred[keep], masks[keep]
+            keep = masks.amax((-2, -1)) > 0  # only keep predictions with masks
+            if not all(keep):  # most predictions have masks
+                pred, masks = pred[keep], masks[keep]  # indexing is slow
         return Results(orig_img, path=img_path, names=self.model.names, boxes=pred[:, :6], masks=masks)

{ultralytics-8.3.214 → ultralytics-8.3.215}/ultralytics/utils/export/__init__.py

@@ -8,6 +8,7 @@ from pathlib import Path
 import torch
 
 from ultralytics.utils import IS_JETSON, LOGGER
+from ultralytics.utils.torch_utils import TORCH_2_4
 
 from .imx import torch2imx  # noqa
 
@@ -36,6 +37,7 @@ def torch2onnx(
     Notes:
         Setting `do_constant_folding=True` may cause issues with DNN inference for torch>=1.12.
     """
+    kwargs = {"dynamo": False} if TORCH_2_4 else {}
     torch.onnx.export(
         torch_model,
         im,
@@ -46,6 +48,7 @@ def torch2onnx(
         input_names=input_names,
         output_names=output_names,
         dynamic_axes=dynamic or None,
+        **kwargs,
     )
 
 

{ultralytics-8.3.214 → ultralytics-8.3.215}/ultralytics/utils/ops.py

@@ -517,12 +517,19 @@ def crop_mask(masks, boxes):
     Returns:
         (torch.Tensor): Cropped masks.
     """
-    _, h, w = masks.shape
-    x1, y1, x2, y2 = torch.chunk(boxes[:, :, None], 4, 1)  # x1 shape(n,1,1)
-    r = torch.arange(w, device=masks.device, dtype=x1.dtype)[None, None, :]  # rows shape(1,1,w)
-    c = torch.arange(h, device=masks.device, dtype=x1.dtype)[None, :, None]  # cols shape(1,h,1)
-
-    return masks * ((r >= x1) * (r < x2) * (c >= y1) * (c < y2))
+    n, h, w = masks.shape
+    if n < 50:  # faster for fewer masks (predict)
+        for i, (x1, y1, x2, y2) in enumerate(boxes.round().int()):
+            masks[i, :y1] = 0
+            masks[i, y2:] = 0
+            masks[i, :, :x1] = 0
+            masks[i, :, x2:] = 0
+        return masks
+    else:  # faster for more masks (val)
+        x1, y1, x2, y2 = torch.chunk(boxes[:, :, None], 4, 1)  # x1 shape(n,1,1)
+        r = torch.arange(w, device=masks.device, dtype=x1.dtype)[None, None, :]  # rows shape(1,1,w)
+        c = torch.arange(h, device=masks.device, dtype=x1.dtype)[None, :, None]  # cols shape(1,h,1)
+        return masks * ((r >= x1) * (r < x2) * (c >= y1) * (c < y2))
 
 
 def process_mask(protos, masks_in, bboxes, shape, upsample: bool = False):
@@ -554,7 +561,7 @@ def process_mask(protos, masks_in, bboxes, shape, upsample: bool = False):
 
     masks = crop_mask(masks, downsampled_bboxes)  # CHW
     if upsample:
-        masks = F.interpolate(masks[None], shape, mode="bilinear", align_corners=False)[0]  # CHW
+        masks = F.interpolate(masks[None], shape, mode="bilinear")[0]  # CHW
     return masks.gt_(0.0)
 
 
@@ -600,7 +607,7 @@ def scale_masks(masks, shape, padding: bool = True):
     top, left = (int(round(pad_h - 0.1)), int(round(pad_w - 0.1))) if padding else (0, 0)
     bottom = mh - int(round(pad_h + 0.1))
     right = mw - int(round(pad_w + 0.1))
-    return F.interpolate(masks[..., top:bottom, left:right], shape, mode="bilinear", align_corners=False)  # NCHW masks
+    return F.interpolate(masks[..., top:bottom, left:right], shape, mode="bilinear")  # NCHW masks
 
 
 def scale_coords(img1_shape, coords, img0_shape, ratio_pad=None, normalize: bool = False, padding: bool = True):

{ultralytics-8.3.214 → ultralytics-8.3.215}/ultralytics.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ultralytics
-Version: 8.3.214
+Version: 8.3.215
 Summary: Ultralytics YOLO 🚀 for SOTA object detection, multi-object tracking, instance segmentation, pose estimation and image classification.
 Author-email: Glenn Jocher <glenn.jocher@ultralytics.com>, Jing Qiu <jing.qiu@ultralytics.com>
 Maintainer-email: Ultralytics <hello@ultralytics.com>