reverting some recently introduced exceptions (#5659)

* reverting some recently introduced exceptions

* Update torchvision/ops/poolers.py

Co-authored-by: Vasilis Vryniotis <[email protected]>

* address PR comments

* replace one more assert with torch._assert:

* address PR comments

* make type checker happy

* Fix bug

* fix bug

* fix for wrong asserts

* attempt to make tests pass

* Fix test_ops tests

* Fix expected exception in tests

* fix typo

* fix tests and format

* fix flake8

* remove one last exception

* fix error

* remove unused immport

* replace fake returns by else

Co-authored-by: Vasilis Vryniotis <[email protected]>

Author: jdsgomes

test/test_models.py

@@ -745,24 +745,24 @@ def test_detection_model_validation(model_fn):
     x = [torch.rand(input_shape)]
 
     # validate that targets are present in training
-    with pytest.raises(ValueError):
+    with pytest.raises(AssertionError):
         model(x)
 
     # validate type
     targets = [{"boxes": 0.0}]
-    with pytest.raises(TypeError):
+    with pytest.raises(AssertionError):
         model(x, targets=targets)
 
     # validate boxes shape
     for boxes in (torch.rand((4,)), torch.rand((1, 5))):
         targets = [{"boxes": boxes}]
-        with pytest.raises(ValueError):
+        with pytest.raises(AssertionError):
             model(x, targets=targets)
 
     # validate that no degenerate boxes are present
     boxes = torch.tensor([[1, 3, 1, 4], [2, 4, 3, 4]])
     targets = [{"boxes": boxes}]
-    with pytest.raises(ValueError):
+    with pytest.raises(AssertionError):
         model(x, targets=targets)
 
 

test/test_models_detection_anchor_utils.py

@@ -16,7 +16,7 @@ def test_incorrect_anchors(self):
         image1 = torch.randn(3, 800, 800)
         image_list = ImageList(image1, [(800, 800)])
         feature_maps = [torch.randn(1, 50)]
-        pytest.raises(ValueError, anc, image_list, feature_maps)
+        pytest.raises(AssertionError, anc, image_list, feature_maps)
 
     def _init_test_anchor_generator(self):
         anchor_sizes = ((10,),)

test/test_ops.py

@@ -138,13 +138,13 @@ def test_autocast(self, x_dtype, rois_dtype):
 
     def _helper_boxes_shape(self, func):
         # test boxes as Tensor[N, 5]
-        with pytest.raises(ValueError):
+        with pytest.raises(AssertionError):
             a = torch.linspace(1, 8 * 8, 8 * 8).reshape(1, 1, 8, 8)
             boxes = torch.tensor([[0, 0, 3, 3]], dtype=a.dtype)
             func(a, boxes, output_size=(2, 2))
 
         # test boxes as List[Tensor[N, 4]]
-        with pytest.raises(ValueError):
+        with pytest.raises(AssertionError):
             a = torch.linspace(1, 8 * 8, 8 * 8).reshape(1, 1, 8, 8)
             boxes = torch.tensor([[0, 0, 3]], dtype=a.dtype)
             ops.roi_pool(a, [boxes], output_size=(2, 2))

torchvision/models/detection/_utils.py

@@ -159,10 +159,14 @@ def encode_single(self, reference_boxes: Tensor, proposals: Tensor) -> Tensor:
         return targets
 
     def decode(self, rel_codes: Tensor, boxes: List[Tensor]) -> Tensor:
-        if not isinstance(boxes, (list, tuple)):
-            raise TypeError(f"This function expects boxes of type list or tuple, instead  got {type(boxes)}")
-        if not isinstance(rel_codes, torch.Tensor):
-            raise TypeError(f"This function expects rel_codes of type torch.Tensor, instead  got {type(rel_codes)}")
+        torch._assert(
+            isinstance(boxes, (list, tuple)),
+            "This function expects boxes of type list or tuple.",
+        )
+        torch._assert(
+            isinstance(rel_codes, torch.Tensor),
+            "This function expects rel_codes of type torch.Tensor.",
+        )
         boxes_per_image = [b.size(0) for b in boxes]
         concat_boxes = torch.cat(boxes, dim=0)
         box_sum = 0
@@ -335,8 +339,7 @@ def __init__(self, high_threshold: float, low_threshold: float, allow_low_qualit
         """
         self.BELOW_LOW_THRESHOLD = -1
         self.BETWEEN_THRESHOLDS = -2
-        if low_threshold > high_threshold:
-            raise ValueError("low_threshold should be <= high_threshold")
+        torch._assert(low_threshold <= high_threshold, "low_threshold should be <= high_threshold")
         self.high_threshold = high_threshold
         self.low_threshold = low_threshold
         self.allow_low_quality_matches = allow_low_quality_matches
@@ -375,8 +378,9 @@ def __call__(self, match_quality_matrix: Tensor) -> Tensor:
 
         if self.allow_low_quality_matches:
             if all_matches is None:
-                raise ValueError("all_matches should not be None")
-            self.set_low_quality_matches_(matches, all_matches, match_quality_matrix)
+                torch._assert(False, "all_matches should not be None")
+            else:
+                self.set_low_quality_matches_(matches, all_matches, match_quality_matrix)
 
         return matches
 

torchvision/models/detection/anchor_utils.py

@@ -84,17 +84,14 @@ def num_anchors_per_location(self):
     def grid_anchors(self, grid_sizes: List[List[int]], strides: List[List[Tensor]]) -> List[Tensor]:
         anchors = []
         cell_anchors = self.cell_anchors
-
-        if cell_anchors is None:
-            ValueError("cell_anchors should not be None")
-
-        if not (len(grid_sizes) == len(strides) == len(cell_anchors)):
-            raise ValueError(
-                "Anchors should be Tuple[Tuple[int]] because each feature "
-                "map could potentially have different sizes and aspect ratios. "
-                "There needs to be a match between the number of "
-                "feature maps passed and the number of sizes / aspect ratios specified."
-            )
+        torch._assert(cell_anchors is not None, "cell_anchors should not be None")
+        torch._assert(
+            len(grid_sizes) == len(strides) == len(cell_anchors),
+            "Anchors should be Tuple[Tuple[int]] because each feature "
+            "map could potentially have different sizes and aspect ratios. "
+            "There needs to be a match between the number of "
+            "feature maps passed and the number of sizes / aspect ratios specified.",
+        )
 
         for size, stride, base_anchors in zip(grid_sizes, strides, cell_anchors):
             grid_height, grid_width = size

torchvision/models/detection/faster_rcnn.py

@@ -1,5 +1,6 @@
 from typing import Any, Optional, Union
 
+import torch
 import torch.nn.functional as F
 from torch import nn
 from torchvision.ops import MultiScaleRoIAlign
@@ -313,10 +314,10 @@ def __init__(self, in_channels, num_classes):
 
     def forward(self, x):
         if x.dim() == 4:
-            if list(x.shape[2:]) != [1, 1]:
-                raise ValueError(
-                    f"x has the wrong shape, expecting the last two dimensions to be [1,1] instead of {list(x.shape[2:])}"
-                )
+            torch._assert(
+                list(x.shape[2:]) == [1, 1],
+                f"x has the wrong shape, expecting the last two dimensions to be [1,1] instead of {list(x.shape[2:])}",
+            )
         x = x.flatten(start_dim=1)
         scores = self.cls_score(x)
         bbox_deltas = self.bbox_pred(x)

torchvision/models/detection/fcos.py

@@ -565,23 +565,25 @@ def forward(
                 like `scores`, `labels` and `mask` (for Mask R-CNN models).
         """
         if self.training:
+
             if targets is None:
-                raise ValueError("In training mode, targets should be passed")
-            for target in targets:
-                boxes = target["boxes"]
-                if isinstance(boxes, torch.Tensor):
-                    if len(boxes.shape) != 2 or boxes.shape[-1] != 4:
-                        raise ValueError(f"Expected target boxes to be a tensor of shape [N, 4], got {boxes.shape}.")
-                else:
-                    raise TypeError(f"Expected target boxes to be of type Tensor, got {type(boxes)}.")
+                torch._assert(False, "targets should not be none when in training mode")
+            else:
+                for target in targets:
+                    boxes = target["boxes"]
+                    torch._assert(isinstance(boxes, torch.Tensor), "Expected target boxes to be of type Tensor.")
+                    torch._assert(
+                        len(boxes.shape) == 2 and boxes.shape[-1] == 4,
+                        f"Expected target boxes to be a tensor of shape [N, 4], got {boxes.shape}.",
+                    )
 
         original_image_sizes: List[Tuple[int, int]] = []
         for img in images:
             val = img.shape[-2:]
-            if len(val) != 2:
-                raise ValueError(
-                    f"expecting the last two dimensions of the Tensor to be H and W instead got {img.shape[-2:]}"
-                )
+            torch._assert(
+                len(val) == 2,
+                f"expecting the last two dimensions of the Tensor to be H and W instead got {img.shape[-2:]}",
+            )
             original_image_sizes.append((val[0], val[1]))
 
         # transform the input
@@ -596,9 +598,9 @@ def forward(
                     # print the first degenerate box
                     bb_idx = torch.where(degenerate_boxes.any(dim=1))[0][0]
                     degen_bb: List[float] = boxes[bb_idx].tolist()
-                    raise ValueError(
-                        "All bounding boxes should have positive height and width."
-                        f" Found invalid box {degen_bb} for target at index {target_idx}."
+                    torch._assert(
+                        False,
+                        f"All bounding boxes should have positive height and width. Found invalid box {degen_bb} for target at index {target_idx}.",
                     )
 
         # get the features from the backbone
@@ -619,11 +621,11 @@ def forward(
         losses = {}
         detections: List[Dict[str, Tensor]] = []
         if self.training:
-            # compute the losses
             if targets is None:
-                raise ValueError("targets should not be none when in training mode")
-
-            losses = self.compute_loss(targets, head_outputs, anchors, num_anchors_per_level)
+                torch._assert(False, "targets should not be none when in training mode")
+            else:
+                # compute the losses
+                losses = self.compute_loss(targets, head_outputs, anchors, num_anchors_per_level)
         else:
             # split outputs per level
             split_head_outputs: Dict[str, List[Tensor]] = {}

torchvision/models/detection/generalized_rcnn.py

@@ -59,23 +59,25 @@ def forward(self, images, targets=None):
         """
         if self.training:
             if targets is None:
-                raise ValueError("In training mode, targets should be passed")
-
-            for target in targets:
-                boxes = target["boxes"]
-                if isinstance(boxes, torch.Tensor):
-                    if len(boxes.shape) != 2 or boxes.shape[-1] != 4:
-                        raise ValueError(f"Expected target boxes to be a tensor of shape [N, 4], got {boxes.shape}.")
-                else:
-                    raise TypeError(f"Expected target boxes to be of type Tensor, got {type(boxes)}.")
+                torch._assert(False, "targets should not be none when in training mode")
+            else:
+                for target in targets:
+                    boxes = target["boxes"]
+                    if isinstance(boxes, torch.Tensor):
+                        torch._assert(
+                            len(boxes.shape) == 2 and boxes.shape[-1] == 4,
+                            f"Expected target boxes to be a tensor of shape [N, 4], got {boxes.shape}.",
+                        )
+                    else:
+                        torch._assert(False, f"Expected target boxes to be of type Tensor, got {type(boxes)}.")
 
         original_image_sizes: List[Tuple[int, int]] = []
         for img in images:
             val = img.shape[-2:]
-            if len(val) != 2:
-                raise ValueError(
-                    f"Expecting the last two dimensions of the input tensor to be H and W, instead got {img.shape[-2:]}"
-                )
+            torch._assert(
+                len(val) == 2,
+                f"expecting the last two dimensions of the Tensor to be H and W instead got {img.shape[-2:]}",
+            )
             original_image_sizes.append((val[0], val[1]))
 
         images, targets = self.transform(images, targets)
@@ -90,9 +92,10 @@ def forward(self, images, targets=None):
                     # print the first degenerate box
                     bb_idx = torch.where(degenerate_boxes.any(dim=1))[0][0]
                     degen_bb: List[float] = boxes[bb_idx].tolist()
-                    raise ValueError(
+                    torch._assert(
+                        False,
                         "All bounding boxes should have positive height and width."
-                        f" Found invalid box {degen_bb} for target at index {target_idx}."
+                        f" Found invalid box {degen_bb} for target at index {target_idx}.",
                     )
 
         features = self.backbone(images.tensors)

torchvision/models/detection/retinanet.py

@@ -494,28 +494,26 @@ def forward(self, images, targets=None):
                 like `scores`, `labels` and `mask` (for Mask R-CNN models).
 
         """
-        if self.training and targets is None:
-            raise ValueError("In training mode, targets should be passed")
-
         if self.training:
             if targets is None:
-                raise ValueError("In training mode, targets should be passed")
-            for target in targets:
-                boxes = target["boxes"]
-                if isinstance(boxes, torch.Tensor):
-                    if len(boxes.shape) != 2 or boxes.shape[-1] != 4:
-                        raise ValueError(f"Expected target boxes to be a tensor of shape [N, 4], got {boxes.shape}.")
-                else:
-                    raise TypeError(f"Expected target boxes to be of type Tensor, got {type(boxes)}.")
+                torch._assert(False, "targets should not be none when in training mode")
+            else:
+                for target in targets:
+                    boxes = target["boxes"]
+                    torch._assert(isinstance(boxes, torch.Tensor), "Expected target boxes to be of type Tensor.")
+                    torch._assert(
+                        len(boxes.shape) == 2 and boxes.shape[-1] == 4,
+                        "Expected target boxes to be a tensor of shape [N, 4].",
+                    )
 
         # get the original image sizes
         original_image_sizes: List[Tuple[int, int]] = []
         for img in images:
             val = img.shape[-2:]
-            if len(val) != 2:
-                raise ValueError(
-                    f"Expecting the two last elements of the input tensors to be H and W instead got {img.shape[-2:]}"
-                )
+            torch._assert(
+                len(val) == 2,
+                f"expecting the last two dimensions of the Tensor to be H and W instead got {img.shape[-2:]}",
+            )
             original_image_sizes.append((val[0], val[1]))
 
         # transform the input
@@ -531,9 +529,10 @@ def forward(self, images, targets=None):
                     # print the first degenerate box
                     bb_idx = torch.where(degenerate_boxes.any(dim=1))[0][0]
                     degen_bb: List[float] = boxes[bb_idx].tolist()
-                    raise ValueError(
+                    torch._assert(
+                        False,
                         "All bounding boxes should have positive height and width."
-                        f" Found invalid box {degen_bb} for target at index {target_idx}."
+                        f" Found invalid box {degen_bb} for target at index {target_idx}.",
                     )
 
         # get the features from the backbone
@@ -554,9 +553,10 @@ def forward(self, images, targets=None):
         detections: List[Dict[str, Tensor]] = []
         if self.training:
             if targets is None:
-                raise ValueError("In training mode, targets should be passed")
-            # compute the losses
-            losses = self.compute_loss(targets, head_outputs, anchors)
+                torch._assert(False, "targets should not be none when in training mode")
+            else:
+                # compute the losses
+                losses = self.compute_loss(targets, head_outputs, anchors)
         else:
             # recover level sizes
             num_anchors_per_level = [x.size(2) * x.size(3) for x in features]