2025-04-17 nightly release (f799a53)

Author: pytorchbot

test/assets/fakedata/draw_boxes_different_label_fill_colors.png

@@ -1384,6 +1384,10 @@ def test_bbox_cxcywhr_to_xyxyxyxy(self):
         box_xyxyxyxy = ops.box_convert(box_tensor, in_fmt="cxcywhr", out_fmt="xyxyxyxy")
         torch.testing.assert_close(box_xyxyxyxy, exp_xyxyxyxy)
 
+        # Reverse conversion
+        box_cxcywhr = ops.box_convert(box_xyxyxyxy, in_fmt="xyxyxyxy", out_fmt="cxcywhr")
+        torch.testing.assert_close(box_cxcywhr, box_tensor)
+
     def test_bbox_xywhr_to_xyxyxyxy(self):
         box_tensor = torch.tensor([[4, 5, 4, 2, 90]], dtype=torch.float)
         exp_xyxyxyxy = torch.tensor([[4, 5, 4, 1, 6, 1, 6, 5]], dtype=torch.float)
@@ -1392,6 +1396,10 @@ def test_bbox_xywhr_to_xyxyxyxy(self):
         box_xyxyxyxy = ops.box_convert(box_tensor, in_fmt="xywhr", out_fmt="xyxyxyxy")
         torch.testing.assert_close(box_xyxyxyxy, exp_xyxyxyxy)
 
+        # Reverse conversion
+        box_xywhr = ops.box_convert(box_xyxyxyxy, in_fmt="xyxyxyxy", out_fmt="xywhr")
+        torch.testing.assert_close(box_xywhr, box_tensor)
+
     @pytest.mark.parametrize("inv_infmt", ["xwyh", "cxwyh", "xwyhr", "cxwyhr", "xxxxyyyy"])
     @pytest.mark.parametrize("inv_outfmt", ["xwcx", "xhwcy", "xwcxr", "xhwcyr", "xyxyxxyy"])
     def test_bbox_invalid(self, inv_infmt, inv_outfmt):

test/test_ops.py

@@ -131,6 +131,23 @@ def test_draw_boxes_with_coloured_labels():
     assert_equal(result, expected)
 
 
+@pytest.mark.skipif(PILLOW_VERSION < (10, 1), reason="The reference image is only valid for PIL >= 10.1")
+def test_draw_boxes_with_coloured_label_backgrounds():
+    img = torch.full((3, 100, 100), 255, dtype=torch.uint8)
+    labels = ["a", "b", "c", "d"]
+    colors = ["green", "#FF00FF", (0, 255, 0), "red"]
+    label_colors = ["green", "red", (0, 255, 0), "#FF00FF"]
+    result = utils.draw_bounding_boxes(
+        img, boxes, labels=labels, colors=colors, fill=True, label_colors=label_colors, fill_labels=True
+    )
+
+    path = os.path.join(
+        os.path.dirname(os.path.abspath(__file__)), "assets", "fakedata", "draw_boxes_different_label_fill_colors.png"
+    )
+    expected = torch.as_tensor(np.array(Image.open(path))).permute(2, 0, 1)
+    assert_equal(result, expected)
+
+
 @pytest.mark.parametrize("fill", [True, False])
 def test_draw_boxes_dtypes(fill):
     img_uint8 = torch.full((3, 100, 100), 255, dtype=torch.uint8)

test/test_utils.py

@@ -178,10 +178,9 @@ def _box_xyxyxyxy_to_xywhr(boxes: Tensor) -> Tensor:
     x1, y1, x3, y3, x2, y2, x4, y4 = boxes.unbind(-1)
     r_rad = torch.atan2(y1 - y3, x3 - x1)
     r = r_rad * 180 / torch.pi
-    cos, sin = torch.cos(r_rad), torch.sin(r_rad)
 
-    w = (x2 - x1) * cos + (y1 - y2) * sin
-    h = (x2 - x1) * sin + (y2 - y1) * cos
+    w = ((x3 - x1) ** 2 + (y1 - y3) ** 2).sqrt()
+    h = ((x3 - x2) ** 2 + (y3 - y2) ** 2).sqrt()
 
     boxes = torch.stack((x1, y1, w, h, r), dim=-1)
 

torchvision/ops/_box_convert.py

@@ -252,13 +252,13 @@ def _xyxyxyxy_to_xywhr(xyxyxyxy: torch.Tensor, inplace: bool) -> torch.Tensor:
         xyxyxyxy = xyxyxyxy.float()
 
     r_rad = torch.atan2(xyxyxyxy[..., 1].sub(xyxyxyxy[..., 3]), xyxyxyxy[..., 2].sub(xyxyxyxy[..., 0]))
-    cos, sin = r_rad.cos(), r_rad.sin()
-    # x1, y1, x3, y3, (x2 - x1), (y2 - y1) x4, y4
-    xyxyxyxy[..., 4:6].sub_(xyxyxyxy[..., :2])
-    # (x2 - x1) * cos + (y1 - y2) * sin = w
-    xyxyxyxy[..., 2] = xyxyxyxy[..., 4].mul(cos).sub(xyxyxyxy[..., 5].mul(sin))
-    # (x2 - x1) * sin + (y2 - y1) * cos = h
-    xyxyxyxy[..., 3] = xyxyxyxy[..., 5].mul(cos).add(xyxyxyxy[..., 4].mul(sin))
+    # x1, y1, (x3 - x1), (y3 - y1), (x2 - x3), (y2 - y3) x4, y4
+    xyxyxyxy[..., 4:6].sub_(xyxyxyxy[..., 2:4])
+    xyxyxyxy[..., 2:4].sub_(xyxyxyxy[..., :2])
+    # sqrt((x3 - x1) ** 2 + (y1 - y3) ** 2) = w
+    xyxyxyxy[..., 2] = xyxyxyxy[..., 2].pow(2).add(xyxyxyxy[..., 3].pow(2)).sqrt()
+    # sqrt((x3 - x2) ** 2 + (y3 - y2) ** 2) = h
+    xyxyxyxy[..., 3] = xyxyxyxy[..., 4].pow(2).add(xyxyxyxy[..., 5].pow(2)).sqrt()
     xyxyxyxy[..., 4] = r_rad.div_(torch.pi).mul_(180.0)
     return xyxyxyxy[..., :5].to(dtype)
 

torchvision/transforms/v2/functional/_meta.py

@@ -162,6 +162,7 @@ def draw_bounding_boxes(
     font: Optional[str] = None,
     font_size: Optional[int] = None,
     label_colors: Optional[Union[List[Union[str, Tuple[int, int, int]]], str, Tuple[int, int, int]]] = None,
+    fill_labels: bool = False,
 ) -> torch.Tensor:
 
     """
@@ -186,7 +187,8 @@ def draw_bounding_boxes(
             `/System/Library/Fonts/` and `~/Library/Fonts/` on macOS.
         font_size (int): The requested font size in points.
         label_colors (color or list of colors, optional): Colors for the label text.  See the description of the
-            `colors` argument for details.  Defaults to the same colors used for the boxes.
+            `colors` argument for details.  Defaults to the same colors used for the boxes, or to black if ``fill_labels`` is True.
+        fill_labels (bool): If `True` fills the label background with specified box color (from the ``colors`` parameter). Default: False.
 
     Returns:
         img (Tensor[C, H, W]): Image Tensor of dtype uint8 with bounding boxes plotted.
@@ -223,8 +225,8 @@ def draw_bounding_boxes(
         )
 
     colors = _parse_colors(colors, num_objects=num_boxes)
-    if label_colors:
-        label_colors = _parse_colors(label_colors, num_objects=num_boxes)  # type: ignore[assignment]
+    if label_colors or fill_labels:
+        label_colors = _parse_colors(label_colors if label_colors else "black", num_objects=num_boxes)  # type: ignore[assignment]
     else:
         label_colors = colors.copy()  # type: ignore[assignment]
 
@@ -259,7 +261,13 @@ def draw_bounding_boxes(
             draw.rectangle(bbox, width=width, outline=color)
 
         if label is not None:
-            margin = width + 1
+            box_margin = 1
+            margin = width + box_margin
+            if fill_labels:
+                left, top, right, bottom = draw.textbbox((bbox[0] + margin, bbox[1] + margin), label, font=txt_font)
+                draw.rectangle(
+                    (left - box_margin, top - box_margin, right + box_margin, bottom + box_margin), fill=color
+                )
             draw.text((bbox[0] + margin, bbox[1] + margin), label, fill=label_color, font=txt_font)  # type: ignore[arg-type]
 
     out = F.pil_to_tensor(img_to_draw)