def __call__(self, image: Image, target: TargetType):
w, h = image.size
image_id = target["image_id"]
image_id = torch.tensor([image_id])
annotations = target.pop("annotations")
annotations = [
obj for obj in annotations if obj.get("iscrowd", 0) == 0
]
boxes = [obj["bbox"] for obj in annotations]
# guard against no boxes via resizing
boxes = torch.as_tensor(boxes, dtype=torch.float32).reshape(-1, 4)
boxes[:, 2:] += boxes[:, :2]
boxes[:, 0::2].clamp_(min=0, max=w)
boxes[:, 1::2].clamp_(min=0, max=h)
classes = [obj["category_id"] for obj in annotations]
classes = torch.tensor(classes, dtype=torch.int64)
segmentations = [obj["segmentation"] for obj in annotations]
masks = convert_segmentation_to_mask(segmentations, h, w)
keypoints = None
if annotations and "keypoints" in annotations[0]:
keypoints = [obj["keypoints"] for obj in annotations]
keypoints = torch.as_tensor(keypoints, dtype=torch.float32)
num_keypoints = keypoints.shape[0]
if num_keypoints:
keypoints = keypoints.view(num_keypoints, -1, 3)
keep = (boxes[:, 3] > boxes[:, 1]) & (boxes[:, 2] > boxes[:, 0])
boxes = boxes[keep]
classes = classes[keep]
masks = masks[keep]
if keypoints is not None:
keypoints = keypoints[keep]
target["boxes"] = boxes
target["labels"] = classes
target["masks"] = masks
target["image_id"] = image_id
if keypoints is not None:
target["keypoints"] = keypoints
# conversion to coco api
area = torch.tensor([obj["area"] for obj in annotations])
iscrowd = torch.tensor([obj.get("iscrowd", 0) for obj in annotations])
target["area"] = area
target["iscrowd"] = iscrowd
return image, target
def __call__(self, image: PILImage.Image, annotation: Dict[str, Any]) -> Tuple[PILImage.Image, PILImage.Image]:
w, h = image.size
segmentations = [obj["segmentation"] for obj in annotation]
cats = [obj["category_id"] for obj in annotation]
if segmentations:
masks = convert_segmentation_to_mask(segmentations, h, w)
cats = torch.as_tensor(cats, dtype=masks.dtype)
# merge all instance masks into a single segmentation map
# with its corresponding categories
target, _ = (masks * cats[:, None, None]).max(dim=0)
# discard overlapping instances
target[masks.sum(0) > 1] = 255
else:
target = torch.zeros((h, w), dtype=torch.uint8)
target = PILImage.fromarray(target.numpy())
return image, target
def __call__(self, image: PILImage.Image, target: TargetType) -> Tuple[PILImage.Image, TargetType]:
w, h = image.size
image_id = target["image_id"]
image_id = torch.tensor([image_id])
annotations = target.pop("annotations")
segmentations = [obj["segmentation"] for obj in annotations]
cats = [obj["category_id"] for obj in annotations]
if segmentations:
masks = convert_segmentation_to_mask(segmentations, h, w)
cats = torch.as_tensor(cats, dtype=masks.dtype)
# merge all instance masks into a single segmentation map
# with its corresponding categories
mask, _ = (masks * cats[:, None, None]).max(dim=0)
# discard overlapping instances
mask[masks.sum(0) > 1] = 255
else:
mask = torch.zeros((h, w), dtype=torch.uint8)
target["mask"] = mask
target["image_id"] = image_id
return image, target