GitLab

* Update ImageReadMode error messages, add newline at the end of image_read_mode.h, replace define with const in image_read_mode.h, add documentation to ImageReadMode enum

* Update readpng_cpu and readjpeg_cpu error messages

* Update image.py documentation

* Moving mobilenet.py to mobilenetv2.py

* Adding mobilenet.py for BC.

* Extending ConvBNReLU for reuse.

* Reduce import scope on mobilenet to only the public and versioned classes and methods.

* Fixing incorrect doc example in MNASNet.

* Fixing incorrect output.

* added the helper method for dimension checks

* unit tests for dimensio check function in functional_tensor

* code formatting and typing

* moved torch image check after tensor check

* unit testcases for test_assert_image_tensor added and refactored

* separate unit testcase file deleted

* assert_image_tensor added to newly created 6 methods

* test cases added for new 6 mthohds

* removed wrongly pasted posterize method and added solarize method for testing
Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>
Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>

* Replacing all torch.jit.annotations with typing

* Replacing remaining typing

* Invert Transform (#3104)

* Adding invert operator.

* Make use of the _assert_channels().

* Update upper bound value.

* Remove private doc from invert, create or reuse generic testing methods to avoid duplication of code in the tests. (#3106)

* Create posterize transformation and refactor common methods to assist reuse. (#3108)

* Implement the solarize transform. (#3112)

* Implement the adjust_sharpness transform (#3114)

* Adding functional operator for sharpness.

* Adding transforms for sharpness.

* Handling tiny images and adding a test.

* Implement the autocontrast transform. (#3117)

* Implement the equalize transform (#3119)

* Implement the equalize transform.

* Turn off deterministic for histogram.

* Fixing test. (#3126)

* Force ratio to be float to avoid numeric overflows on blend. (#3127)

* Separate the tests of Adjust Sharpness from ColorJitter. (#3128)

* Add AutoAugment Policies and main Transform (#3142)

* Separate the tests of Adjust Sharpness from ColorJitter.

* Initial implementation, not-jitable.

* AutoAugment passing JIT.

* Adding tests/docs, changing formatting.

* Update test.

* Fix formats

* Fix documentation and imports.

* Apply changes from code review:
- Move the transformations outside of AutoAugment on a separate method.
- Renamed degenerate method for sharpness for better clarity.

* Update torchvision/transforms/functional.py
Co-authored-by: vfdev <vfdev.5@gmail.com>

* Apply more changes from code review:
- Add InterpolationMode parameter.
- Move all declarations away from AutoAugment constructor and into the private method.

* Update documentation.

* Apply suggestions from code review
Co-authored-by: Francisco Massa <fvsmassa@gmail.com>

* Apply changes from code review:
- Refactor code to eliminate as any to() and clamp() as possible.
- Reuse methods where possible.
- Apply speed ups.

* Replacing pad.
Co-authored-by: vfdev <vfdev.5@gmail.com>
Co-authored-by: Francisco Massa <fvsmassa@gmail.com>

* Removing VISION_API from backward() methods and adding a ops.h

* Fixing clang format.

* Fix an issue that ShuffleNetV2 model is exported to a wrong ONNX file if dynamic_axes field was provided.

* Add a ut for the bug fix.

* Fix flake8 issue.

* Don't access each element in x.shape, use x.size() instead.
Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>

* Adapt to new torch export API for dictionary

Co-authored-by: Francisco Massa <fvsmassa@gmail.com>

* Enable ONNX test in circle CI

Summary:
* Reduce unnecessary header inclusions in models and io.

* Move autocast to separate folder and hide autograd implementation in an anonymous namespace.

* Moving files in subfolders.

Reviewed By: fmassa

Differential Revision: D25461523

fbshipit-source-id: 756eeb6848aacaa474de4825ed4c1045d17e2cea

* [WIP] Added CONTRIBUTING.md

* Updated CONTRIBUTING.md

* Update

* Moving deform_conv2d op registration.

* Moving nms op registration.

* Moving new_empty_tensor op registration.

* Moving ps_roi_align op registration.

* Moving ps_roi_pool op registration.

* Moving roi_align op registration.

* Moving roi_pool op registration.

* Restoring headers for forward/backward and fixing styles.

* Restoring the test hack on windows.

* Stricter header inclusion.

* Remove TravisCI

Add hub tests to CircleCI, coverage and ONNX are still missing

* Install torchvision dependencies on CI

The previous names were incompatible with internal fbcode naming rules

This has been fixed in PyTorch in https://github.com/pytorch/pytorch/pull/40897

Makes it appease fbcode internal tests

accimage always stores images as uint8, so let's be compatible with the internal representation. Tests were failing without this as it would return a float32 image normalized between 0-1

We could have errors such as aten/src/ATen/native/cpu/PowKernel.cpp:41:5:  runtime error: 5.7896e+76 is outside the range of representable values of type 'float'

It's not used currently and can lead to exceptions such as runtime error: 5.7896e+76 is outside the range of representable values of type 'float'

* Fixing upperbound value on tests and documentation.

* Limit the number of channels on adjust_* transoforms.

* Encapsulate and standardize deform_conv2d (#3074)

* Rename files.

* Standardizing method names.

* Adding anonymous namespaces.

* Applying C++ naming rules and alinging variable names across headers and cpp files.

* Syncing names across implementations.

* Rename deform_conv2d.h to deform_conv2d.cpp

* Use header files:
- Create header files for kernel implementation and remove definitions from vision_*.h files.
- Eliminate unnecessary headers and ensure all cpp include their headers.

* Change the naming convention for kernel implementations.

* Remove the _param postfix from the variables and standardizing names.

* Exposing public forward/backward methods to the C++ API and moving methods around to minimize git blame changes.

* Encapsulate and standardize nms (#3081)

* Syncing, where possible, the names of functions across devices.

* Adding all internal functions in anonymous namespaces.

* Renaming C++/CUDA kernel files and moving operator code from header to cpp file.

* Create foreach cpp file a separate header file with "public" functions.

* Removing unnecessary repeated includes.

* Update CMakeLists.txt to include all headers.

* Encapsulate and standardize ps_roi_align (#3082)

* Renaming C++ files & methods according to recommended naming conventions and aligning them with Python's API.
Syncing, where possible, the names of functions across devices.

* Adding all internal functions in anonymous namespaces.

* Renaming C++/CUDA kernel files and moving operator code from header to cpp file.

* Create foreach cpp file a separate header file with "public" functions.

* Removing unnecessary repeated includes.

* Encapsulate and standardize ps_roi_pool (#3084)

* Renaming C++ files & methods according to recommended naming conventions and aligning them with Python's API.

* Adding all internal functions in anonymous namespaces.

* Renaming C++/CUDA kernel files and moving operator code from header to cpp file.

* Create foreach cpp file a separate header file with "public" functions.

* Removing unnecessary repeated includes.

* Encapsulate and standardize roi_align (#3085)

* Renaming C++ files & methods according to recommended naming conventions and aligning them with Python's API.

* Adding all internal functions in anonymous namespaces.

* Renaming C++/CUDA kernel files and moving operator code from header to cpp file.

* Create foreach cpp file a separate header file with "public" functions.

* Removing unnecessary repeated includes.

* Encapsulate and standardize roi_pool  (#3088)

* Renaming C++ files & methods according to recommended naming conventions and aligning them with Python's API.

* Adding all internal functions in anonymous namespaces.

* Syncing variable names between the cpp files and their header files.

* Renaming C++/CUDA kernel files and moving operator code from header to cpp file.

* Create foreach cpp file a separate header file with "public" functions.

* Removing unnecessary repeated includes.

* Encapsulate and standardize new_empty_tensor_op (#3089)

* Renaming C++ files & methods according to recommended naming conventions and aligning them with Python's API.

* Create foreach cpp file a separate header file with "public" functions.

* Adding all internal functions in anonymous namespaces.

* Convert to const ref all possible parameters.

* Removing unnecessary repeated includes.

* Encapsulate and standardize C++ Ops - Clean up (#3094)

* Removing unnecessary repeated includes.

* Remove unnecessary vision_cpu.h, vision_cuda.h, autocast.h.

* Fixing naming convention and correcting method names on macros.

* Turn on clang formatter for cu files and fixing broken styles.

* Replace "#ifndef ... #define ... #endif" with "#pragma once" on header files.

* Adding operator methods in vision::ops namespace. (#3096)

* Adding operator methods in vision::ops namespace.

* Replace general.h with macros.h

* Adding vision.h to the necessary cpp files.

Replace tabs with spaces, add newlines to files and replace whitelist with allowlist

For now this simplifies a few things for fbcode integration. We can add this back in the future

Operator registration of functions with noexcept doesn't work on some compilers, see https://github.com/pytorch/pytorch/issues/48667

* Fill color support for tensor affine transforms

* PEP fix

* Docstring changes and float support

* Docstring update for transforms and float type cast

* Cast only for Tensor

* Temporary patch for lack of Union type support, plus an extra unit test

* More plausible bilinear filling for tensors

* Keep things simple & New docstrings

* Fix lint and other issues after merge

* make it in one line

* Docstring and some code modifications

* More tests and corresponding changes for transoforms and docstring changes

* Simplify test configs

* Update test_functional_tensor.py

* Update test_functional_tensor.py

* Move assertions
Co-authored-by: vfdev <vfdev.5@gmail.com>

* add UUID in LOG() in decoder

* Fix lint

* More lint

* concatenate small tensors into big ones to reduce the use of shared file descriptor (#1694)

Summary:
Pull Request resolved: https://github.com/pytorch/vision/pull/1694



- PT dataloader forks worker process to speed up the fetching of dataset example.  The recommended way of multiprocess context is `forkserver` rather than `fork`.

- Main process and worker processes will share the dataset class instance, which avoid duplicating the dataset and save memory. In this process, `ForkPickler(..).dumps(...)` will be called to serialize the objects, including objects within dataset instance recursively. `VideoClips` instance internally uses O(N) `torch.Tensor` to store per-video information, such as pts, and possible clips, where N is the No. of videos.

- During dumping, each `torch.Tensor` will use one File Descriptor (FD). The OS default max limit of FD is 65K by using `ulimit -n` to query. The number of tensors in `VideoClips` often exceeds the limit.

- To resolve this issue, we use a few big tensors by concatenating small tensors in the `__getstate__()` method, which will be called during pickling. This will only require O(1) tensors.

- When this diff is landed, we can abondon D19173248

In D19173397, in ClassyVision, we change the mp context from `fork` to `forkserver`, and finally can run the PT dataloader without hanging issues.

Reviewed By: fmassa

Differential Revision: D19179991

fbshipit-source-id: c8716775c7c154aa33d93b25d112d2a59ea688a9

* Try to fix Windows

* Try fix Windows v2

* Disable tests on Windows

* Add back necessary part

* Try fix OSX (and maybe Windows)

* Fix

* Try enabling Windows
Co-authored-by: Zhicheng Yan <zyan3@fb.com>