Dispatch style box_area and box_iou

Author: alperenunlu

docs/source/ops.rst

@@ -50,10 +50,8 @@ These utility functions perform various operations on bounding boxes.
     :template: function.rst
 
     box_area
-    box_area_center
     box_convert
     box_iou
-    box_iou_center
     clip_boxes_to_image
     complete_box_iou
     distance_box_iou

test/test_ops.py

@@ -1418,9 +1418,9 @@ def test_bbox_convert_jit(self):
         torch.testing.assert_close(scripted_cxcywh, box_cxcywh)
 
 
-class TestBoxArea:
+class TestBoxAreaXYXY:
     def area_check(self, box, expected, atol=1e-4):
-        out = ops.box_area(box)
+        out = ops.box_area(box, fmt="xyxy")
         torch.testing.assert_close(out, expected, rtol=0.0, check_dtype=False, atol=atol)
 
     @pytest.mark.parametrize("dtype", [torch.int8, torch.int16, torch.int32, torch.int64])
@@ -1445,15 +1445,15 @@ def test_float16_box(self):
 
     def test_box_area_jit(self):
         box_tensor = torch.tensor([[0, 0, 100, 100], [0, 0, 0, 0]], dtype=torch.float)
-        expected = ops.box_area(box_tensor)
+        expected = ops.box_area(box_tensor, fmt="xyxy")
         scripted_fn = torch.jit.script(ops.box_area)
         scripted_area = scripted_fn(box_tensor)
         torch.testing.assert_close(scripted_area, expected)
 
 
-class TestBoxAreaCenter:
+class TestBoxAreaCXCYWH:
     def area_check(self, box, expected, atol=1e-4):
-        out = ops.box_area_center(box)
+        out = ops.box_area(box, fmt="cxcywh")
         torch.testing.assert_close(out, expected, rtol=0.0, check_dtype=False, atol=atol)
 
     @pytest.mark.parametrize("dtype", [torch.int8, torch.int16, torch.int32, torch.int64])
@@ -1480,9 +1480,9 @@ def test_float16_box(self):
     def test_box_area_jit(self):
         box_tensor = ops.box_convert(torch.tensor([[0, 0, 100, 100], [0, 0, 0, 0]], dtype=torch.float),
                                      in_fmt="xyxy", out_fmt="cxcywh")
-        expected = ops.box_area_center(box_tensor)
-        scripted_fn = torch.jit.script(ops.box_area_center)
-        scripted_area = scripted_fn(box_tensor)
+        expected = ops.box_area(box_tensor, fmt="cxcywh")
+        scripted_fn = torch.jit.script(ops.box_area)
+        scripted_area = scripted_fn(box_tensor, fmt="cxcywh")
         torch.testing.assert_close(scripted_area, expected)
 
 
@@ -1509,22 +1509,22 @@ def gen_box(size, dtype=torch.float):
     return torch.cat([xy1, xy2], axis=-1)
 
 
-class TestIouBase:
+class TestIouXYXYBase:
     @staticmethod
     def _run_test(target_fn: Callable, actual_box1, actual_box2, dtypes, atol, expected):
         for dtype in dtypes:
             actual_box1 = torch.tensor(actual_box1, dtype=dtype)
             actual_box2 = torch.tensor(actual_box2, dtype=dtype)
             expected_box = torch.tensor(expected)
-            out = target_fn(actual_box1, actual_box2)
+            out = target_fn(actual_box1, actual_box2, fmt="xyxy")
             torch.testing.assert_close(out, expected_box, rtol=0.0, check_dtype=False, atol=atol)
 
     @staticmethod
     def _run_jit_test(target_fn: Callable, actual_box: List):
         box_tensor = torch.tensor(actual_box, dtype=torch.float)
-        expected = target_fn(box_tensor, box_tensor)
+        expected = target_fn(box_tensor, box_tensor, fmt="xyxy")
         scripted_fn = torch.jit.script(target_fn)
-        scripted_out = scripted_fn(box_tensor, box_tensor)
+        scripted_out = scripted_fn(box_tensor, box_tensor, fmt="xyxy")
         torch.testing.assert_close(scripted_out, expected)
 
     @staticmethod
@@ -1534,19 +1534,19 @@ def _cartesian_product(boxes1, boxes2, target_fn: Callable):
         result = torch.zeros((N, M))
         for i in range(N):
             for j in range(M):
-                result[i, j] = target_fn(boxes1[i].unsqueeze(0), boxes2[j].unsqueeze(0))
+                result[i, j] = target_fn(boxes1[i].unsqueeze(0), boxes2[j].unsqueeze(0), fmt="xyxy")
         return result
 
     @staticmethod
     def _run_cartesian_test(target_fn: Callable):
         boxes1 = gen_box(5)
         boxes2 = gen_box(7)
-        a = TestIouBase._cartesian_product(boxes1, boxes2, target_fn)
-        b = target_fn(boxes1, boxes2)
+        a = TestIouXYXYBase._cartesian_product(boxes1, boxes2, target_fn)
+        b = target_fn(boxes1, boxes2, fmt="xyxy")
         torch.testing.assert_close(a, b)
 
 
-class TestBoxIou(TestIouBase):
+class TestBoxIouXYXY(TestIouXYXYBase):
     int_expected = [[1.0, 0.25, 0.0], [0.25, 1.0, 0.0], [0.0, 0.0, 1.0], [0.0625, 0.25, 0.0]]
     float_expected = [[1.0, 0.9933, 0.9673], [0.9933, 1.0, 0.9737], [0.9673, 0.9737, 1.0]]
 
@@ -1568,22 +1568,22 @@ def test_iou_cartesian(self):
         self._run_cartesian_test(ops.box_iou)
 
 
-class TestIouCenterBase:
+class TestIouCXCYWHBase:
     @staticmethod
     def _run_test(target_fn: Callable, actual_box1, actual_box2, dtypes, atol, expected):
         for dtype in dtypes:
             actual_box1 = torch.tensor(actual_box1, dtype=dtype)
             actual_box2 = torch.tensor(actual_box2, dtype=dtype)
             expected_box = torch.tensor(expected)
-            out = target_fn(actual_box1, actual_box2)
+            out = target_fn(actual_box1, actual_box2, fmt="cxcywh")
             torch.testing.assert_close(out, expected_box, rtol=0.0, check_dtype=False, atol=atol)
 
     @staticmethod
     def _run_jit_test(target_fn: Callable, actual_box: List):
         box_tensor = torch.tensor(actual_box, dtype=torch.float)
-        expected = target_fn(box_tensor, box_tensor)
+        expected = target_fn(box_tensor, box_tensor, fmt="cxcywh")
         scripted_fn = torch.jit.script(target_fn)
-        scripted_out = scripted_fn(box_tensor, box_tensor)
+        scripted_out = scripted_fn(box_tensor, box_tensor, fmt="cxcywh")
         torch.testing.assert_close(scripted_out, expected)
 
     @staticmethod
@@ -1593,19 +1593,19 @@ def _cartesian_product(boxes1, boxes2, target_fn: Callable):
         result = torch.zeros((N, M))
         for i in range(N):
             for j in range(M):
-                result[i, j] = target_fn(boxes1[i].unsqueeze(0), boxes2[j].unsqueeze(0))
+                result[i, j] = target_fn(boxes1[i].unsqueeze(0), boxes2[j].unsqueeze(0), fmt="cxcywh")
         return result
 
     @staticmethod
     def _run_cartesian_test(target_fn: Callable):
         boxes1 = ops.box_convert(gen_box(5), in_fmt="xyxy", out_fmt="cxcywh")
         boxes2 = ops.box_convert(gen_box(7), in_fmt="xyxy", out_fmt="cxcywh")
-        a = TestIouCenterBase._cartesian_product(boxes1, boxes2, target_fn)
-        b = target_fn(boxes1, boxes2)
+        a = TestIouCXCYWHBase._cartesian_product(boxes1, boxes2, target_fn)
+        b = target_fn(boxes1, boxes2, fmt="cxcywh")
         torch.testing.assert_close(a, b)
 
 
-class TestBoxIouCenter(TestIouBase):
+class TestBoxIouCXCYWH(TestIouCXCYWHBase):
     int_expected = [[1.0, 0.25, 0.0], [0.25, 1.0, 0.0], [0.0, 0.0, 1.0], [0.04, 0.16, 0.0]]
     float_expected = [[1.0, 0.9933, 0.9673], [0.9933, 1.0, 0.9737], [0.9673, 0.9737, 1.0]]
 
@@ -1618,13 +1618,49 @@ class TestBoxIouCenter(TestIouBase):
         ],
     )
     def test_iou(self, actual_box1, actual_box2, dtypes, atol, expected):
-        self._run_test(ops.box_iou_center, actual_box1, actual_box2, dtypes, atol, expected)
+        self._run_test(ops.box_iou, actual_box1, actual_box2, dtypes, atol, expected)
 
     def test_iou_jit(self):
-        self._run_jit_test(ops.box_iou_center, INT_BOXES_CXCYWH)
+        self._run_jit_test(ops.box_iou, INT_BOXES_CXCYWH)
 
     def test_iou_cartesian(self):
-        self._run_cartesian_test(ops.box_iou_center)
+        self._run_cartesian_test(ops.box_iou)
+
+class TestIouBase:
+    @staticmethod
+    def _run_test(target_fn: Callable, actual_box1, actual_box2, dtypes, atol, expected):
+        for dtype in dtypes:
+            actual_box1 = torch.tensor(actual_box1, dtype=dtype)
+            actual_box2 = torch.tensor(actual_box2, dtype=dtype)
+            expected_box = torch.tensor(expected)
+            out = target_fn(actual_box1, actual_box2)
+            torch.testing.assert_close(out, expected_box, rtol=0.0, check_dtype=False, atol=atol)
+
+    @staticmethod
+    def _run_jit_test(target_fn: Callable, actual_box: List):
+        box_tensor = torch.tensor(actual_box, dtype=torch.float)
+        expected = target_fn(box_tensor, box_tensor)
+        scripted_fn = torch.jit.script(target_fn)
+        scripted_out = scripted_fn(box_tensor, box_tensor)
+        torch.testing.assert_close(scripted_out, expected)
+
+    @staticmethod
+    def _cartesian_product(boxes1, boxes2, target_fn: Callable):
+        N = boxes1.size(0)
+        M = boxes2.size(0)
+        result = torch.zeros((N, M))
+        for i in range(N):
+            for j in range(M):
+                result[i, j] = target_fn(boxes1[i].unsqueeze(0), boxes2[j].unsqueeze(0))
+        return result
+
+    @staticmethod
+    def _run_cartesian_test(target_fn: Callable):
+        boxes1 = gen_box(5)
+        boxes2 = gen_box(7)
+        a = TestIouBase._cartesian_product(boxes1, boxes2, target_fn)
+        b = target_fn(boxes1, boxes2)
+        torch.testing.assert_close(a, b)
 
 
 class TestGeneralizedBoxIou(TestIouBase):

torchvision/ops/__init__.py

@@ -2,10 +2,8 @@
 from .boxes import (
     batched_nms,
     box_area,
-    box_area_center,
     box_convert,
     box_iou,
-    box_iou_center,
     clip_boxes_to_image,
     complete_box_iou,
     distance_box_iou,
@@ -42,9 +40,7 @@
     "clip_boxes_to_image",
     "box_convert",
     "box_area",
-    "box_area_center",
     "box_iou",
-    "box_iou_center",
     "generalized_box_iou",
     "distance_box_iou",
     "complete_box_iou",

torchvision/ops/boxes.py

@@ -272,7 +272,30 @@ def box_convert(boxes: Tensor, in_fmt: str, out_fmt: str) -> Tensor:
     return boxes
 
 
-def box_area(boxes: Tensor) -> Tensor:
+def box_area(boxes: Tensor, fmt: str = "xyxy") -> Tensor:
+    """
+    Computes the area of a set of bounding boxes from a given format.
+
+    Args:
+        boxes (Tensor[N, 4]): boxes for which the area will be computed. They
+            are expected to be in (x1, y1, x2, y2) format with
+            ``0 <= x1 < x2`` and ``0 <= y1 < y2``.
+        fmt (str): Format of given boxes. Supported formats are ['xyxy', 'cxcywh']. Default: "xyxy"
+
+    Returns:
+        Tensor[N]: the area for each box
+    """
+    if fmt == "xyxy":
+        boxes = box_area_xyxy(boxes=boxes)
+    elif fmt == "cxcywh":
+        boxes = box_area_cxcywh(boxes=boxes)
+    else:
+        raise ValueError(f"Unsupported Box Area Calculation for given fmt {fmt}")
+
+    return boxes
+
+
+def box_area_xyxy(boxes: Tensor) -> Tensor:
     """
     Computes the area of a set of bounding boxes, which are specified by their
     (x1, y1, x2, y2) coordinates.
@@ -286,12 +309,12 @@ def box_area(boxes: Tensor) -> Tensor:
         Tensor[N]: the area for each box
     """
     if not torch.jit.is_scripting() and not torch.jit.is_tracing():
-        _log_api_usage_once(box_area)
+        _log_api_usage_once(box_area_xyxy)
     boxes = _upcast(boxes)
     return (boxes[:, 2] - boxes[:, 0]) * (boxes[:, 3] - boxes[:, 1])
 
 
-def box_area_center(boxes: Tensor) -> Tensor:
+def box_area_cxcywh(boxes: Tensor) -> Tensor:
     """
     Computes the area of a set of bounding boxes, which are specified by their
     (cx, cy, w, h) coordinates.
@@ -305,16 +328,39 @@ def box_area_center(boxes: Tensor) -> Tensor:
         Tensor[N]: the area for each box
     """
     if not torch.jit.is_scripting() and not torch.jit.is_tracing():
-        _log_api_usage_once(box_area_center)
+        _log_api_usage_once(box_area_cxcywh)
     boxes = _upcast(boxes)
     return boxes[:, 2] * boxes[:, 3]
 
 
+def box_iou(boxes1: Tensor, boxes2: Tensor, fmt: str = "xyxy") -> Tensor:
+    """
+    Return intersection-over-union (Jaccard index) between two sets of boxes from a given format.
+
+    Args:
+        boxes1 (Tensor[N, 4]): first set of boxes
+        boxes2 (Tensor[M, 4]): second set of boxes
+        fmt (str): Format of given boxes. Supported formats are ['xyxy', 'cxcywh']. Default: "xyxy"
+
+
+    Returns:
+        Tensor[N, M]: the NxM matrix containing the pairwise IoU values for every element in boxes1 and boxes2
+    """
+    if fmt == "xyxy":
+        iou = box_iou_xyxy(boxes1=boxes1, boxes2=boxes2)
+    elif fmt == "cxcywh":
+        iou = box_iou_cxcywh(boxes1=boxes1, boxes2=boxes2)
+    else:
+        raise ValueError(f"Unsupported Box IoU Calculation for given fmt {fmt}")
+
+    return iou
+
+
 # implementation from https://github.com/kuangliu/torchcv/blob/master/torchcv/utils/box.py
 # with slight modifications
-def _box_inter_union(boxes1: Tensor, boxes2: Tensor) -> Tuple[Tensor, Tensor]:
-    area1 = box_area(boxes1)
-    area2 = box_area(boxes2)
+def _box_inter_union_xyxy(boxes1: Tensor, boxes2: Tensor) -> Tuple[Tensor, Tensor]:
+    area1 = box_area(boxes1, fmt="xyxy")
+    area2 = box_area(boxes2, fmt="xyxy")
 
     lt = torch.max(boxes1[:, None, :2], boxes2[:, :2])  # [N,M,2]
     rb = torch.min(boxes1[:, None, 2:], boxes2[:, 2:])  # [N,M,2]
@@ -327,7 +373,7 @@ def _box_inter_union(boxes1: Tensor, boxes2: Tensor) -> Tuple[Tensor, Tensor]:
     return inter, union
 
 
-def box_iou(boxes1: Tensor, boxes2: Tensor) -> Tensor:
+def box_iou_xyxy(boxes1: Tensor, boxes2: Tensor) -> Tensor:
     """
     Return intersection-over-union (Jaccard index) between two sets of boxes.
 
@@ -342,15 +388,15 @@ def box_iou(boxes1: Tensor, boxes2: Tensor) -> Tensor:
         Tensor[N, M]: the NxM matrix containing the pairwise IoU values for every element in boxes1 and boxes2
     """
     if not torch.jit.is_scripting() and not torch.jit.is_tracing():
-        _log_api_usage_once(box_iou)
-    inter, union = _box_inter_union(boxes1, boxes2)
+        _log_api_usage_once(box_iou_xyxy)
+    inter, union = _box_inter_union_xyxy(boxes1, boxes2)
     iou = inter / union
     return iou
 
 
-def _box_inter_union_center(boxes1: Tensor, boxes2: Tensor) -> Tuple[Tensor, Tensor]:
-    area1 = box_area_center(boxes1)
-    area2 = box_area_center(boxes2)
+def _box_inter_union_cxcywh(boxes1: Tensor, boxes2: Tensor) -> Tuple[Tensor, Tensor]:
+    area1 = box_area(boxes1, fmt="cxcywh")
+    area2 = box_area(boxes2, fmt="cxcywh")
 
     lt = torch.max(boxes1[:, None, :2] - boxes1[:, None, 2:] / 2, boxes2[:, :2] - boxes2[:, 2:] / 2)  # [N,M,2]
     rb = torch.min(boxes1[:, None, :2] + boxes1[:, None, 2:] / 2, boxes2[:, :2] + boxes2[:, 2:] / 2)  # [N,M,2]
@@ -363,7 +409,7 @@ def _box_inter_union_center(boxes1: Tensor, boxes2: Tensor) -> Tuple[Tensor, Ten
     return inter, union
 
 
-def box_iou_center(boxes1: Tensor, boxes2: Tensor) -> Tensor:
+def box_iou_cxcywh(boxes1: Tensor, boxes2: Tensor) -> Tensor:
     """
     Return intersection-over-union (Jaccard index) between two sets of boxes.
 
@@ -378,8 +424,8 @@ def box_iou_center(boxes1: Tensor, boxes2: Tensor) -> Tensor:
         Tensor[N, M]: the NxM matrix containing the pairwise IoU values for every element in boxes1 and boxes2
     """
     if not torch.jit.is_scripting() and not torch.jit.is_tracing():
-        _log_api_usage_once(box_iou_center)
-    inter, union = _box_inter_union_center(boxes1, boxes2)
+        _log_api_usage_once(box_iou_cxcywh)
+    inter, union = _box_inter_union_cxcywh(boxes1, boxes2)
     iou = inter / union
     return iou
 
@@ -403,7 +449,7 @@ def generalized_box_iou(boxes1: Tensor, boxes2: Tensor) -> Tensor:
     if not torch.jit.is_scripting() and not torch.jit.is_tracing():
         _log_api_usage_once(generalized_box_iou)
 
-    inter, union = _box_inter_union(boxes1, boxes2)
+    inter, union = _box_inter_union_xyxy(boxes1, boxes2)
     iou = inter / union
 
     lti = torch.min(boxes1[:, None, :2], boxes2[:, :2])