Summary: 90 instances, 67 unique

Text	Count
# TODO: (eellison) T54974082 https://github.com/pytorch/pytorch/issues/26744/pytorch/issues/26744	1
# TODO add some warnings in this case	1
# FIXME this is kind of a hack, but we will jump to the previous keyframe	1
// TODO: reset params.formats	1
# TODO: should this handle containers?	1
// TODO: consider supporting PNG_INFO_tRNS	1
# TODO check if stream needs to always be the video stream here or not	1
// TODO: There should be an easier way to do this	5
// FIXME: change this when batches of size > 1 are allowed	1
# TODO format that into engineering format	1
# TODO change this	1
# TODO: replace this method with torch.iinfo when it gets torchscript support.	1
# TODO https://github.com/pytorch/vision/pull/4232#pullrequestreview-730461659	11
# TODO: simplify when indexing without rank will be supported by ONNX	1
# TODO: https://github.com/pytorch/pytorch/issues/26792	1
# TODO: deprecate eventually	3
// TODO here torch::relu_ and torch::adaptive_avg_pool2d wrapped in	1
# FIXME: Is this needed? SqueezeNet should only be called from the	1
# FIXME: Building torchvision with ffmpeg on MacOS or with Python 3.9	1
# TODO: Revert it once the bug is fixed.	1
# TODO: Also save the optimizer and scheduler	1
// FIXME: This is private, API might change:	1
# TODO: we should expect bincount to always be faster than histc, but this	1
// TODO: maybe we can specify dynamic shared memory size before calling the	1
# TODO raise a warning?	1
# FIXME remove this and make paste_masks_in_image run on the GPU	1
# TODO: support test	1
# FIXME: causes crash. See the following GitHub issues for more details.	1
# TODO: Do we want a list or a dict?	1
// TODO: Once torch::from_file handles UTF-8 paths correctly, we should move	1
# TODO: replace with dtype.is_floating_point when torchscript supports it	2
# TODO: make this more precise	1
# TODO: refactor with utils.verify_str_arg	1
# TODO: There might be a way to vectorize this	1
// TODO: maybe we can specify dynamic shared memory size before calling the	1
# TODO use pretrained as a string to specify the backend	5
# TODO: resume, pretrained, and weights should be in an exclusive arg group	1
# TODO: https://github.com/pytorch/pytorch/issues/26731	1
# FIXME We don't know if we should expect this to happen	1
# TODO: Ideally, this should be part of the eval transforms preset, instead	1
# See FIXME above	1
# TODO: create a segmentation feature	1
// FIXME: Remove this section once we can use at::native for android xplat	1
# FIXME: https://github.com/pytorch/vision/issues/3367	1
# FIXME: This checking is not done for the other models	1
#  TODO: Benchmark this against the approach described at https://github.com/pytorch/vision/pull/5197#discussion_r786251298	1
# TODO: Move this to a function	2
# TODO: might be weird to not take `num_workers` into account	1
# TODO: make this more informative	1
# FIXME need to take into account that the datasets	2
# TODO add a warning	1
# FIXME: https://github.com/pytorch/pytorch/issues/65000	1
# TODO format that into engineering format	1
# TODO: this needs to be more elaborate	1
# FIXME: squeezenet1_x() functions	1
# FIXME: Eliminate copy-pasted code for fill standardization and _augmentation_space() by moving stuff on a base class	1
# FIXME assume for now that testing uses the largest scale	1
# TODO: specify the return type	1
# TODO: this is weird as it drops more elements than it should	1
# FIXME: pad input in case it is smaller than crop_box	1
// TODO: add read from memory option	1
# TODO: vectorize this instead of using a for loop	1
# TODO: this needs to be revisited to allow subclassing of custom transforms	1
# TODO: Once torchscript supports Enums with staticmethod	1
# TODO : replace below with a dynamic padding when support is added in ONNX	1
# TODO: Remove this warning after ONNX opset 16 is supported.	1
# TODO add back the assert	1