GitLab

* Add support of different color types in readpng.

* Adding test images and unit-tests.

* Use closest possible type.

* Fix formatting.

* Add modulation input for DeformConv2D

* lint

* Patch for GPU CI

* Remove bad cache on CI

* Make read_file accept unicode strings on Windows

* More fixes

* Remove definitions from source files

* Move string definitions to header

* Add checks

* Fix comments

* Update macro

* Fix comments

* Fix lint

* include windows header

* Change func signature in header

* Use from_blob

* Fix fread calls

* Fix clang format

* Fix missing return

* Avoid copy

* Making all model constructors explicit.

* formatting.

* Fixing types.

* Dispatcher + Autocast.

* + Autograd.

* Formating.

* Clean up and refactor PSROIPool implementation:
- Remove primitive const declaration from method names.
- Using references when possible.
- Fix variable naming.

* Restore include headers.

* New line at end of file.

* Resolving conflict, final cleanup, ordering method consistently across files.

* Fixing types.

* Dispatcher + Autocast.

* + Autograd.

* Formating.

* Fixing return casting with autocast.

* Clean up and refactor ROIPool implementation:
- Remove primitive const declaration from method names.
- Using references when possible.

* Restore include headers.

* New line at end of file.

* Clean up and refactor ROIAlign implementation:
- Remove primitive const declaration from method names.
- Passing as const ref instead of value where possible.
- Remove unnecessary headers.

* Adding back include for cpu.

* Restore include headers.

* Fixing types.

* Dispatcher + Autocast.

* + Autograd.

* Clean up and refactor PSROIAlign implementation:
- Remove primitive const declaration from method names.
- Using references when possible.
- Sync naming of internal methods with other ops.

* Restoring names of internal methods to avoid conflicts.

* Restore include headers.

* Clean up and refactor DeformConv implementation:
- Remove primitive const declaration from method names.
- Passing as const ref instead of value where possible.
- Aligning method names between cpu and cuda.

* Adding newline.

* Adding back include for cpu.

* Restoring method names of private methods to avoid conflicts.

* Restore include headers.

* Clean up and refactor ROIAlign implementation:
- Remove primitive const declaration from method names.
- Remove unnecessary headers.
- Aligning method names between cpu and cuda.

* Adding back include for cpu.

* Restoring method names of private methods to avoid conflicts.

* Restore include headers.

- Replaced tuple creation by one acceptable by majority of compilers

* Splitting tuples of stride, padding and dilation of DeformConv.

* Fixing types.

* Dispatcher + Autocast.

* + Autograd.

* Moving contiguous() convertions away dispatcher and into the implementations.

* Removing rvalue references.

* Minor refactoring based on static analysis on the code of models:
- Convert unnecessary value parameter to constant reference.
- Use move to avoid unnecessary copies.
- Eliminate unused include.

* Replace moves with const references.

* Fixing formatting.

* Remove explicit declaration on constructors.

- Remove unnecessary return.

- Replace deprecated string.h with cstring.
- Switch to nullptr.

* Convert to const reference and eliminate unnecessary bool casting.

* Removing unnecessary namespace use.

* Ensure torchvision operators are registered in C++ via weak symbols

* Add note to README on how to ensure that torchvision operators are available in C++

* Fix dllimport/dllexport on windows, format files

* Factor out common macros in single file

* Expose cuda_version in the API, use it to avoid pruning of ops initializer

* [DEBUG] Check wheel relocation issues

* Call delocate on Mac

* Use yum instead of conda

* Do not copy ffmpeg dylibs

* Linux dry run

* Do not download FFmpeg on Mac

* Install bzip2

* Restore FFmpeg on Windows

* Remove ffmpeg temporarily

* Do not remove dependencies

* Disable FFmpeg temporarily on Linux wheels

* Test relocation in Linux

* Add docstring

* Call relocation script

* Minor error correction

* Import auditwheel only on Linux

* Restore ffmpeg again on Windows

* Return *device

* Try fix Windows

* Fix clang-format

* Start windows patchwork

* Copy all DLLs

* Disable FFmpeg on Windows for now

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

* Added rois shape check in C++

* Fixes code formatting

* Remove accidental include

* - Updated code according to the review
- Replaced old AT_ASSERT/ERROR by new TORCH_CHECK

* Adding checks on forward and backward passes.

* Adding unit-tests.

* Improve read_file for Windows

* Fix lint

* Use _stat64 instead

* Apply suggesion

* Fix lint

* adding base files

* setup modification to actually build the thing

* video api constructor registration

* FAIL metadata

* FAIL update for QS

* revert

* debugging with Victor

* adding base files

* setup modification to actually build the thing

* video api constructor registration

* FAIL metadata

* FAIL update for QS

* revert

* debugging with Victor

* metadata registration works

* API build next

* test

* Merge change

* formatting parameters to avoid the segfault

* next now works on a video

* make size of the output tensor format dependent

* Make next work on audio stream only as well

* refactoring the _setCurrentStream param

* Fixing the last frame return and sensor

* todo docs

* Formatting

* cleanup and comments

* introducing new tests for the API

* cleanup

* Comment out unnecesary format (will add following FFMPEG fix)

* Reformat parsing function

* removing the seek bug `get_decoder_params`

* Removing unnecessary code/variables

* enforce RGB24 as a reading format (will crash before ffmpeg fix)

* permute the dimensions to return (RGB x H x W)

* Changing the return type to std::tuple<torch::Tensor, double> as opposed to tensor list

* Adjusting tests for the new return type

* remove unnecessary jitter

* clangangangang

* Metadata return changes (#1)

* remove implicit calls to set a current stream (#2)

* Adding new tests to check the accuracy of the seek

* cleanup debugging statements

* adding base files

* setup modification to actually build the thing

* video api constructor registration

* FAIL metadata

* FAIL update for QS

* revert

* debugging with Victor

* adding base files

* video api constructor registration

* FAIL metadata

* FAIL update for QS

* revert

* debugging with Victor

* metadata registration works

* API build next

* test

* Merge change

* formatting parameters to avoid the segfault

* next now works on a video

* make size of the output tensor format dependent

* Make next work on audio stream only as well

* refactoring the _setCurrentStream param

* Fixing the last frame return and sensor

* todo docs

* Formatting

* cleanup and comments

* introducing new tests for the API

* cleanup

* Comment out unnecesary format (will add following FFMPEG fix)

* Reformat parsing function

* removing the seek bug `get_decoder_params`

* Removing unnecessary code/variables

* enforce RGB24 as a reading format (will crash before ffmpeg fix)

* permute the dimensions to return (RGB x H x W)

* Changing the return type to std::tuple<torch::Tensor, double> as opposed to tensor list

* Adjusting tests for the new return type

* remove unnecessary jitter

* clangangangang

* Metadata return changes (#1)

* remove implicit calls to set a current stream (#2)

* Adding new tests to check the accuracy of the seek

* cleanup debugging statements

* Addressing PR comments

* addressing Francisco's comments

* CLANG build formatting

* Updated testing to test against pyav for the video tensor reads

* Formatting

* remove pyav from pip deps and add it to conda build

* add pyav and ffmeped to conda builds

* Formatting?

* Setting up linter once and for all hopefully

* Testing pyav

* Fix to 8.0.0

* Try 6.2.0

* See what happens with av from pip

* Remove FFMPEG blocker

* What is going on?

* More tests

* Forgot something

* unblocker

* Check if cache is messing up with things

* Now try with different ffmpeg

* Now try with different ffmpeg

* Testing pyav

* Fix to 8.0.0

* Try 6.2.0

* See what happens with av from pip

* What is going on?

* More tests

* Forgot something

* Check if cache is messing up with things

* Now try with different ffmpeg

* Now try with different ffmpeg

* Do not install av

* Test with ffmpeg 4.2

* clean up video tests

* cleaning up the tests a bit to better test partial reading

* arrgh linter

* Forgot the av test

* forgot av test

* checkout build files from master

* revert circleci

* addressing Franciscos comments

* addressing Franciscos comments

* Ignore ffmpeg in travis
Co-authored-by: Francisco Massa <fvsmassa@gmail.com>
Co-authored-by: Edgar Andrés Margffoy Tuay <andfoy@gmail.com>

* Add write_file

* Fix lint

* Add PyTorch FFmpeg to wheel and conda distributions

* Try to install wget from conda

* Add yq flag on Mac

* Correct copy instructions

* Use cURL on Windows

* Call bzip2 directly due to msys2/MSYS2-packages#1548

* Copy ffmpeg binaries to system-wide directories

* Try to use std:c++17 on Windows

* Try to define ssize_t on Windows

* Use C++14

* Declare AVRational structs explicitly

* Initialize AVRational explicitly

* Replace macro to prevent errors on Windows

* Replace AV_TIME_BASE_Q

* Add library paths for video extension

* Force ffmpeg from pytorch channels?

* Fix clang style warnings

* Update CONDA_CHANNEL_FLAGS

* Fix clang style issues

* Update unittest

* Use FFmpeg 4.2

* Install correct version on Mac

* Pin av version to 8.0.0

* Fix string formatting issue

* Fix pip pinning

* Try with 8.0.1

* Use av 8.0.2

* Remove trailling whitespaces

* Disable test_io_opt.py

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

* Add read_file

* Add test for non-existent file

* Fix lint

* Lint v2

* Try fix windows

* Try fix Windows v2

* Lint

* Windows v3

* Missed one change in the adapted function

* Try again on Windows

* One more try on Windows

* Give up on tempfile for now

* Are extensions what's missing on Windows?

* Investigating if the issue is on our side

* Try deleting tensor which could hold into the file

* Put back temporary folder

* Add encode/write_png functions

* Do not redefine

* Style issues correction

* Comply with low-level interface

* Minor comment correction

* Add python frontend functions

* Add encode_png test

* Pass compession level to encode_png

* Do not compare output buffers

* Convert to bytes

* Compare pil image instead of buffer

* Add error tests

* Add test_write_png

* Remove png test assets

* Register writePNG correctly

* Update write_png docstring

* Do not preserve PIL image beyond the scope
Co-authored-by: Francisco Massa <fvsmassa@gmail.com>

* Add decode_image op

* Fix lint

* More lint

* Add C10_EXPORT

* Add decode and write JPEG ops

* Fix styling issues

* Use int64_t instead of int

* Use std::string

* Use jpegcommon.h for read_jpeg

* Minor updates to error handling in read

* Include header only once

* Reverse header inclusion

* Update common header

* Add common definitions

* Include string

* Include header?

* Include header?

* Add Python frontend calls

* Use unsigned long directly

* Fix style issues

* Include cstddef

* Ignore clang-format on cstddef

* Also include stdio

* Add JPEG and PNG include dirs

* Use C10_EXPORT

* Add JPEG encoding test

* Set quality to 75 by default and add write jpeg test

* Minor error correction

* Use assertEquals by assertEqual

* Remove test results

* Use pre-saved PIL output

* Remove extra PIL call

* Use read_jpeg instead of PIL

* Add error tests

* Address review comments

* Fix style issues

* Set test case to uint8

* Update test error check

* Apply suggestions from code review

* Fix clang-format

* Fix lint

* Fix test

* Remove unused file

* Fix regex error message

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

* Make JPEG/PNG return images in CHW format

* Use int array

* Add CMake build pipeline

* Add CMake build workflow

* Add executable permissions to script

* Install cmake on Windows/MacOS

* Install conda-build before setting up MSVC

* Install PyTorch from nightly

* Do not use conda-build variables

* Add path to CMake

* Install libpng and libjpeg

* Perform make

* Call msbuild on Windows

* Add missing yq

* Use vc_env_helper

* Use string instruction

* Escape configuration option

* Remove configuration flag

* Try to pass -p

* Use caret to escape equal sign

* Escape string option in Windows

* Try to call other bat

* Remove Windows/GPU CMake

* Add tracing cpp test

* Script model instead of tracing it

* Try to register operators manually

* Use manylinux-cuda102

* Activate conda env on Linux

* Build and run sample tracing test

* Add empty echo

* Remove unnecessary register

* Copy headers on Mac

* Revert to 2xlarge

* Include /usr/local/include on Mac

* Install pillow on Windows

* Install future

* Install torchvision on Windows

* Set include flag

* Add torchlib to PATH

* Normalize path via cygpath

* Register ops on Windows

* Minor error correction

* Register CPU/GPU ops on DLL library and register ops via reference

* Install dataclasses

* Install dataclasses using pip

* Address clang formatting issue

* Try to use an actual GPU instance on Linux

* Remove extra environment section

* Declare environment explicitly

* Regenerate

* Pass env variables to Dokcer

* Regenerate circleci

* Test tracing on GPU

* Use GPU medium

* Regenerate

* Use cuda101

* Regenerate

* Do not use pre-trained weights

Avoids having to download pretrained files, which could cause flaky tests
Co-authored-by: Francisco Massa <fvsmassa@gmail.com>

* add autocasting on ROCm

* enable ROIAlign autocasting on ROCm

* enable NMS autocasting on ROCm

* fix to use correct torch CUDA APIs

* [WIP] Fixes #2598
- Adjusted num_kernels and batch_size according to kMaxGridNum * CUDA_NUM_THREADS
- Tests to add

* - Redefined kMaxGridNum as max grid according to current CUDA device
- Added test to check the code from issue and compared grads CPU/CUDA

* Fixed static kMaxGridNum evaluation to dynamic

* Add corruption cases

* Read jpeg headers until exhaustion

* Minor error correction

* Add test script

* Raise exception when image is truncated

* Add test

* Skip damaged_jpeg folder

* Compare against basename

* Remove unused test file

* Fix wrong clamping in RoIAlign with aligned=True

* Fix silly mistake

* Bugfix pointed out during code-review

* Fixes Xiao's repro

* Ports nms to use full dispatcher

* Move HIPGuard to nms_cuda

* clang-format

* run models in test_models.py on GPU if available

* Francisco's comment, also disable cuda model tests to see if CPU alone still passes

* cuda tests now pass locally, although still not comparing to saved numerics

* add note for thing to ask francisco

* Allow cuda and cpu tests to share a data file

* ignore suffix if unneeded

* Skip autocast numerics checks for a few models

* Add roi_align test
Co-authored-by: Michael Carilli <mcarilli@nvidia.com>