GitLab

Summary:
Allow writes of >= 2^32 bytes. High-res video can cross this threshold sometimes.
LHS is `size_t`, but RHS is all `int32`, and will overflow for output tensors >2Gb.

Reviewed By: jsawruk

Differential Revision: D21255664

fbshipit-source-id: 7b4c5da989777297a89e73615aaeee8c7a13186a
Co-authored-by: Tilak Sharma <tilaksharma@fb.com>

* Make RandomHorizontalFlip torchscriptable

* Make _is_tensor_a_torch_image more generic

* Make RandomVerticalFlip torchscriptable (#2283)

* Make RandomVerticalFlip torchscriptable

* Fix lint

* Bugfix in pad

* Address review comments

* Fix lint

Summary:

We've made two recent changes to QAT in PyTorch core:
1. add support for SyncBatchNorm
2. make eager mode QAT prepare scripts respect device affinity

This PR updates the torchvision QAT reference script to take
advantage of both of these.  This should be landed after
https://github.com/pytorch/pytorch/pull/39337 (the last PT
fix) to avoid compatibility issues.

Test Plan:

```
python -m torch.distributed.launch
  --nproc_per_node 8
  --use_env
  references/classification/train_quantization.py
  --data-path {imagenet1k_subset}
  --output-dir {tmp}
  --sync-bn
```

Reviewers:

Subscribers:

Tasks:

Tags:

* Add more tests to NMS

* Fix lint

* Remove interpolate in favor of PyTorch's implementation

* Bugfix

* Bugfix

* Fix binary builds with CUDA 9.2 on Windows

* Remove packages according to VC_YEAR

* get pts directly instead of storing full frames to get pts later

* fix linting

* add initial pts value
sort pts

* catch decoding errors for read_video_timestamp

* Add norm_layer to MobileNetV2

* Add simple test case

* Small fix

Co-authored-by: eellison <eellison@fb.com>

* add compatibility table to README

* Update README.rst

* Update README.rst
Co-authored-by: Francisco Massa <fvsmassa@gmail.com>

* Avoid `using` in header files

* Fix clang_format

* use clang-format-7 to reformat code

* add layer freezing param to maskrcnn_resnet50_fpn

* freeze ayer param in keypointrcnn_resnet50_fpn

* layer freeze tests for mask and keypoint rcnn

* correct linting errors

* correct linting errors.

* correct linting errors

* Fix #2221, DenseNet issue with gradient checkpoints (memory_efficient=True)

* Add grad/param count test for mem_efficient densenet

* Deprecate Conv2d, ConvTranspose2d and BatchNorm

* Fix lint

* Update README.md

added some clarity to get the examples executable.  Waiting to hear back if instructions should mention to setup COCO dataset

* Update README.md

* Update README.md

* Update README.md

* Update README.md

* Update README.md

* Fixing nms on boxes when no detection

* test

* Fix for scale_factor computation

* remove newline

* Fix for mask_rcnn dynanmic axes

* Clean up

* Update transform.py

* Fix for torchscript

* Fix scripting errors

* Fix annotation

* Fix lint

* Fix annotation

* Fix for interpolate scripting

* Fix for scripting

* refactoring

* refactor the code

* Fix annotation

* Fixed annotations

* Added test for resize

* lint

* format

* bump ORT

* ort-nightly version

* Going to ort 1.1.0

* remove version

* install typing-extension

* Export model for images with no detection

* Upgrade ort nightly

* update ORT

* Update test_onnx.py

* updated tests

* Updated tests

* merge

* Update transforms.py

* Update cityscapes.py

* Update celeba.py

* Update caltech.py

* Update pkg_helpers.bash

* Clean up

* Clean up for dynamic split

* Remove extra casts

* flake8

* Fix for mask rcnn no detection export

* clean up

* Enable mask rcnn tests

* Added test

* update ORT

* Update .travis.yml

* fix annotation

* Clean up roi_heads

* clean up

* clean up misc ops

Another instance of integer division using the division operator. In this case line 266 already shows the correct formulation, so line 185 only needs the update.

* get pts directly instead of storing full frames to get pts later

* fix linting

* add initial pts value
sort pts

Integer division using the div operator is deprecated and will throw a RuntimeError in PyTorch 1.6 (and on PyTorch Master very soon). Running a test build with a recent Torchvision commit and integer division using div ('/') disabled revealed this integer division. 

I'll re-run the tests once this is fixed in case it's masking additional issues.

* Make copy of targets in GeneralizedRCNNTransform

* Fix flake8

Co-authored-by: eellison <eellison@fb.com>

Summary:

Redo of https://github.com/pytorch/vision/pull/2191

Makes the classification QAT tutorial not crash when used
with DDP. There were two issues:

1. the model was moved to GPU before the observers were added, and they
are created on CPU. In the context of this repo, the fix is to finalize
the model before moving to GPU. We can potentially follow up with a
better error message in the future, in a separate PR.
2. the QAT conversion was running on the DDP'ed model, which had various
problems. The fix is to unwrap the model from DDP before cloning it for
evaluation.

There is still work to do on verifying that BN is working correctly in
QAT + DDP, but saving that for a separate PR.

Test Plan:

```
python -m torch.distributed.launch --use_env references/classification/train_quantization.py --data-path {path_to_imagenet_1k} --output_dir {output_dir}
```

Reviewers:

Subscribers:

Tasks:

Tags:

* Adds as_tensor to functional.py

Similar functionality to to_tensor without the default conversion to float and division by 255.
Also adds support for Image mode 'L'.

* Adds tests to AsTensor()

Adds tests to AsTensor and removes the conversion to float and division by 255.

* Adds AsTensor to transforms.py

Calls the as_tensor function in functionals and adds the function AsTensor as callable from transforms.

* Removes the pic.mode == 'L'

This was handled by the else condition previously so I'll remove it.

* Fix Lint issue

Adds two line breaks between functions to fix lint issue

* Replace from_numpy with as_tensor

Removes the extra if conditionals and replaces from_numpy with as_tensor.

* Renames as_tensor to pil_to_tensor

Renames the function as_tensor to pil_to_tensor and narrows the scope of the function.  At the same time also creates a flag that defaults to True for swapping to the channels first format.

* Renames AsTensor to PILToImage

Renames the function AsTensor to PILToImage and modifies the description.  Adds the swap_to_channelsfirst boolean variable to indicate if the user wishes to change the shape of the input.

* Add the __init__ function to PILToTensor 

Add the __init__ function to PILToTensor since it contains the swap_to_channelsfirst parameter now.

* fix lint issue

remove trailing white space

* Fix the tests

Reflects the name change to PILToTensor and the parameter to the function as well as the new narrowed scope that the function only accepts PIL images.

* fix tests

Instead of undoing the transpose just create a new tensor and test that one.

* Add the view back

Add img.view(pic.size[1], pic.size[0], len(pic.getbands())) back to outside the if condition.

* fix test

fix conversion from torch tensor to PIL back to torch tensor.

* fix lint issues

* fix lint

remove trailing white space

* Fixed the channel swapping tensor test

Torch tranpose operates differently than numpy transpose.  Changed operation to permute.

* Add mode='F'

Add mode information when converting to PIL Image from Float Tensor.

* Added inline comments to follow shape changes

* ToPILImage converts FloatTensors to uint8

* Remove testing not swapping

* Removes the swap_channelsfirst parameter

Makes the channel swapping the default behavior.

* Remove the swap_channelsfirst argument

Remove the swap_channelsfirst argument and makes the swapping the default functionality.

As requested by @fmassa in comment
https://github.com/pytorch/vision/issues/2216#issuecomment-630166106
from #2216.

* Fix missing include for OSX in video decoder

* clang-format

* freeze layers only if pretrained backbone is used

If pretrained backbone is not used and one intends to train the entire network from scratch, no layers should be frozen.

* function argument to control the trainable features

Depending on the size of dataset one might want to control the number of tunable parameters in the backbone, and this parameter in hyper parameter optimization for the dataset. It would be nice to have this function support this.

* ensuring tunable layer argument is valid

* backbone freezing in fasterrcnn_resnet50_fpn

Handle backbone freezing in fasterrcnn_resnet50_fpn function rather than the resnet_fpn_backbone function that it uses to get the backbone.

* remove layer freezing code

layer freezing code has been moved to fasterrcnn_resnet50_fpn function that consumes resnet_fpn_backbone function.

* correcting linting errors

* correcting linting errors

* move freezing logic to resnet_fpn_backbone

Moved layer freezing logic to resnet_fpn_backbone with an additional parameter.

* remove layer freezing from fasterrcnn_resnet50_fpn

Layer freezing logic has been moved to resnet_fpn_backbone. This function only ensures that the all layers are made trainable if pretrained models are not used.

* update example resnet_fpn_backbone docs

* correct typo in var name

* correct indentation

* adding test case for layer freezing in faster rcnn

This PR adds functionality to specify the number of trainable layers while initializing the faster rcnn using fasterrcnn_resnet50_fpn function. This commits adds a test case to test this functionality.

* updating layer freezing condition for clarity

More information in PR

* remove linting errors

* removing linting errors

* removing linting errors

Fixes https://github.com/pytorch/vision/issues/2214#issuecomment-628636663

I don't know why the building is not working with `--expt-relaxed-constexpr` flag set, but it is generally a good idea to declare this as `__host__ __device__`

* Check target boxes input on generalized_rcnn.py

* Fix target box validation in generalized_rcnn.py

* Add tests for input validation of detection models