Remove support of targets on presets.

Author: datumbox

gallery/plot_optical_flow.py

@@ -96,7 +96,7 @@ def plot(imgs, **imshow_kwargs):
 def preprocess(img1_batch, img2_batch):
     img1_batch = F.resize(img1_batch, size=[520, 960])
     img2_batch = F.resize(img2_batch, size=[520, 960])
-    return transforms(img1_batch, img2_batch)[:2]
+    return transforms(img1_batch, img2_batch)
 
 
 img1_batch, img2_batch = preprocess(img1_batch, img2_batch)

gallery/plot_repurposing_annotations.py

@@ -146,11 +146,8 @@ def show(imgs):
 print(img.size())
 
 tranforms = weights.transforms()
-img, _ = tranforms(img)
-target = {}
-target["boxes"] = boxes
-target["labels"] = labels = torch.ones((masks.size(0),), dtype=torch.int64)
-detection_outputs = model(img.unsqueeze(0), [target])
+img = tranforms(img)
+detection_outputs = model(img.unsqueeze(0))
 
 
 ####################################

gallery/plot_visualization_utils.py

@@ -81,7 +81,7 @@ def show(imgs):
 weights = FasterRCNN_ResNet50_FPN_Weights.DEFAULT
 transforms = weights.transforms()
 
-batch, _ = transforms(batch_int)
+batch = transforms(batch_int)
 
 model = fasterrcnn_resnet50_fpn(weights=weights, progress=False)
 model = model.eval()
@@ -131,7 +131,7 @@ def show(imgs):
 model = fcn_resnet50(weights=weights, progress=False)
 model = model.eval()
 
-normalized_batch, _ = transforms(batch)
+normalized_batch = transforms(batch)
 output = model(normalized_batch)['out']
 print(output.shape, output.min().item(), output.max().item())
 
@@ -272,7 +272,7 @@ def show(imgs):
 weights = MaskRCNN_ResNet50_FPN_Weights.DEFAULT
 transforms = weights.transforms()
 
-batch, _ = transforms(batch_int)
+batch = transforms(batch_int)
 
 model = maskrcnn_resnet50_fpn(weights=weights, progress=False)
 model = model.eval()
@@ -397,7 +397,7 @@ def show(imgs):
 weights = KeypointRCNN_ResNet50_FPN_Weights.DEFAULT
 transforms = weights.transforms()
 
-person_float, _ = transforms(person_int)
+person_float = transforms(person_int)
 
 model = keypointrcnn_resnet50_fpn(weights=weights, progress=False)
 model = model.eval()

torchvision/transforms/_presets.py

@@ -20,12 +20,10 @@
 
 
 class ObjectDetectionEval(nn.Module):
-    def forward(
-        self, img: Tensor, target: Optional[Dict[str, Tensor]] = None
-    ) -> Tuple[Tensor, Optional[Dict[str, Tensor]]]:
+    def forward(self, img: Tensor) -> Tensor:
         if not isinstance(img, Tensor):
             img = F.pil_to_tensor(img)
-        return F.convert_image_dtype(img, torch.float), target
+        return F.convert_image_dtype(img, torch.float)
 
 
 class ImageClassificationEval(nn.Module):
@@ -95,28 +93,22 @@ def __init__(
         self._interpolation = interpolation
         self._interpolation_target = interpolation_target
 
-    def forward(self, img: Tensor, target: Optional[Tensor] = None) -> Tuple[Tensor, Optional[Tensor]]:
+    def forward(self, img: Tensor) -> Tensor:
         if isinstance(self._size, list):
             img = F.resize(img, self._size, interpolation=self._interpolation)
         if not isinstance(img, Tensor):
             img = F.pil_to_tensor(img)
         img = F.convert_image_dtype(img, torch.float)
         img = F.normalize(img, mean=self._mean, std=self._std)
-        if target:
-            if isinstance(self._size, list):
-                target = F.resize(target, self._size, interpolation=self._interpolation_target)
-            if not isinstance(target, Tensor):
-                target = F.pil_to_tensor(target)
-            target = target.squeeze(0).to(torch.int64)
-        return img, target
+        return img
 
 
 class OpticalFlowEval(nn.Module):
-    def forward(
-        self, img1: Tensor, img2: Tensor, flow: Optional[Tensor] = None, valid_flow_mask: Optional[Tensor] = None
-    ) -> Tuple[Tensor, Tensor, Optional[Tensor], Optional[Tensor]]:
-
-        img1, img2, flow, valid_flow_mask = self._pil_or_numpy_to_tensor(img1, img2, flow, valid_flow_mask)
+    def forward(self, img1: Tensor, img2: Tensor) -> Tuple[Tensor, Tensor]:
+        if not isinstance(img1, Tensor):
+            img1 = F.pil_to_tensor(img1)
+        if not isinstance(img2, Tensor):
+            img2 = F.pil_to_tensor(img2)
 
         img1 = F.convert_image_dtype(img1, torch.float32)
         img2 = F.convert_image_dtype(img2, torch.float32)
@@ -128,19 +120,4 @@ def forward(
         img1 = img1.contiguous()
         img2 = img2.contiguous()
 
-        return img1, img2, flow, valid_flow_mask
-
-    def _pil_or_numpy_to_tensor(
-        self, img1: Tensor, img2: Tensor, flow: Optional[Tensor], valid_flow_mask: Optional[Tensor]
-    ) -> Tuple[Tensor, Tensor, Optional[Tensor], Optional[Tensor]]:
-        if not isinstance(img1, Tensor):
-            img1 = F.pil_to_tensor(img1)
-        if not isinstance(img2, Tensor):
-            img2 = F.pil_to_tensor(img2)
-
-        if flow is not None and not isinstance(flow, Tensor):
-            flow = torch.from_numpy(flow)
-        if valid_flow_mask is not None and not isinstance(valid_flow_mask, Tensor):
-            valid_flow_mask = torch.from_numpy(valid_flow_mask)
-
-        return img1, img2, flow, valid_flow_mask
+        return img1, img2