GitLab

* [0.4_video] models - initial commit

* addressing fmassas inline comments

* pep8 and flake8

* simplify "hacks"

* sorting out latest comments

* nitpick

* Updated tests and constructors

* Added docstrings - ready to merge

* in_channels_stage2 from backbone.inplanes

* remove type for backward compatible

Fix the old flatten method which use the size(0) to caculate the batch size, the old method will intruduce Gather opertion in the onnx output, which will faild parsed by tensorRT 5.0 (#1134)

updated all docstrings and code references for boxes to be consistent with the scheme (x1, y1, x2, y2) (#1110)

* add float32 to keypoint_rcnn docs

* add float32 to faster_rcnn docs

* add float32 to mask_rcnn

* Update faster_rcnn.py

* Update keypoint_rcnn.py

* Update mask_rcnn.py

* Update faster_rcnn.py

* make keypoints float

* make masks uint8

* Update keypoint_rcnn.py

* make labels Int64

* make labels Int64

* make labels Int64

* Add checks for boxes, labels, masks, keypoints

* update mask dim

* remove dtype

* check only if targets is not None

* account for targets being a list

* update target to be list of dict

* Update faster_rcnn.py

* Update keypoint_rcnn.py

* allow boxes to be of float16 type as well

* remove checks on mask

* Add memory_efficient args to densenet docs

This PR adds the arg `memory_efficient` to densenet docs and also mentions the relevant publication.

* linting

This PR updates the docs of MnasNet

Fixes #1047.

* Add paper references to VGG

* Add paper references to ResNet and its variants

* fix transform for rcnns so original images are unchanged

* transform does not change input list anymore

transform does not change input list anymore. Improve code according to reviewer comment

* transform for maskrcnn no longer modify input list

transform for maskrcnn no longer modify input list. improve code according to comment

* transform for maskrcnn no longer modifies input list

* Fixed width multiplier

Layer channels are now rounded to a multiple of 8, as per the official tensorflow implementation.  I found this fix when looking through: https://github.com/d-li14/mobilenetv2.pytorch

* Channel multiple now a user configurable option

The official tensorflow slim mobilenet v2 implementation rounds the number of channels in each layer to a multiple of 8.  This is now user configurable - 1 turns off rounding

* Fixed whitespace error

Fixed error: ./torchvision/models/mobilenet.py:152:1: W293 blank line contains whitespace

* add wide resnet

* add docstring for wide resnet

* update WRN-50-2 model

* add docs

* extend WRN docstring

* use pytorch storage for WRN

* fix rebase

* fix typo in docs

* Add initial mnasnet impl

* Remove all type hints, comply with PyTorch overall style

* Expose models

* Remove avgpool from features() and add separately

* Fix python3-only stuff, replace subclasses with functions

* fix __all__

* Fix typo

* Remove conditional dropout

* Make dropout functional

* Addressing @fmassa's feedback, round 1

* Replaced adaptive avgpool with mean on H and W to prevent collapsing the batch dimension

* Partially address feedback

* YAPF

* Removed redundant class vars

* Update urls to releases

* Add information to models.rst

* Replace init with kaiming_normal_ in fan-out mode

* Use load_state_dict_from_url

I grepped the repo for Ouputs and these were the only occurences

This PR uses a protected method for loading and initializing the segmentation models. Relevant #875

* GPU efficient Densenets

* removed `import math`

* Changed 'efficient' to 'memory_efficient'

* Add tests

* Bugfix in test

* Fix lint

* Remove unecessary formatting

* allow user to define residual settings

* 4spaces

* linting errors

* backward compatible, and added test

Add 'import torch' in example

Consider the difference of the division operator between Python 2.x and Python 3.x.

This makes it consistent with the other models, which returns nouns in plurial

Also adds documentation for the segmentation models

* Add more documentation for the ops

* Add documentation for Faster R-CNN

* Add documentation for Mask R-CNN and Keypoint R-CNN

* Improve doc for RPN

* Add basic doc for GeneralizedRCNNTransform

* Lint fixes

Those were not free parameters, and can be inferred via the size of the output feature map

* Add COCO pre-trained weights for Faster R-CNN R-50 FPN

* Add weights for Mask R-CNN and Keypoint R-CNN

* Split mask_rcnn.py into several files

* Lint

* Move segmentation models to its own folder

* Add missing files

* Remove dependency from functool in ShuffleNetsV2

This PR removes the dependence of the ShuffleNetV2 code from `functool`.

* flake fix

* [Remove] Use stride in 1x1 in resnet

This is temporary

* Move files to torchvision

Inference works

* Now seems to give same results

Was using the wrong number of total iterations in the end...

* Distributed evaluation seems to work

* Factor out transforms into its own file

* Enabling horizontal flips

* MultiStepLR and preparing for launches

* Add warmup

* Clip gt boxes to images

Seems to be crucial to avoid divergence. Also reduces the losses over different processes for better logging

* Single-GPU batch-size 1 of CocoEvaluator works

* Multi-GPU CocoEvaluator works

Gives the exact same results as the other one, and also supports batch size > 1

* Silence prints from pycocotools

* Commenting unneeded code for run

* Fixes

* Improvements and cleanups

* Remove scales from Pooler

It was not a free parameter, and depended only on the feature map dimensions

* Cleanups

* More cleanups

* Add misc ops and totally remove maskrcnn_benchmark

* nit

* Move Pooler to ops

* Make FPN slightly more generic

* Minor improvements or FPN

* Move FPN to ops

* Move functions to utils

* Lint fixes

* More lint

* Minor cleanups

* Add FasterRCNN

* Remove modifications to resnet

* Fixes for Python2

* More lint fixes

* Add aspect ratio grouping

* Move functions around

* Make evaluation use all images for mAP, even those without annotations

* Bugfix with DDP introduced in last commit

* [Check] Remove category mapping

* Lint

* Make GroupedBatchSampler prioritize largest clusters in the end of iteration

* Bugfix for selecting the iou_types during evaluation

Also switch to using the torchvision normalization now on, given that we are using torchvision base models

* More lint

* Add barrier after init_process_group

Better be safe than sorry

* Make evaluation only use one CPU thread per process

When doing multi-gpu evaluation, paste_masks_in_image is multithreaded and throttles evaluation altogether. Also change default for aspect ratio group to match Detectron

* Fix bug in GroupedBatchSampler

After the first epoch, the number of batch elements could be larger than batch_size, because they got accumulated from the previous iteration. Fix this and also rename some variables for more clarity

* Start adding KeypointRCNN

Currently runs and perform inference, need to do full training

* Remove use of opencv in keypoint inference

PyTorch 1.1 adds support for bicubic interpolation which matches opencv (except for empty boxes, where one of the dimensions is 1, but that's fine)

* Remove Masker

Towards having mask postprocessing done inside the model

* Bugfixes in previous change plus cleanups

* Preparing to run keypoint training

* Zero initialize bias for mask heads

* Minor improvements on print

* Towards moving resize to model

Also remove class mapping specific to COCO

* Remove zero init in bias for mask head

Checking if it decreased accuracy

* [CHECK] See if this change brings back expected accuracy

* Cleanups on model and training script

* Remove BatchCollator

* Some cleanups in coco_eval

* Move postprocess to transform

* Revert back scaling and start adding conversion to coco api

The scaling didn't seem to matter

* Use decorator instead of context manager in evaluate

* Move training and evaluation functions to a separate file

Also adds support for obtaining a coco API object from our dataset

* Remove unused code

* Update location of lr_scheduler

Its behavior has changed in PyTorch 1.1

* Remove debug code

* Typo

* Bugfix

* Move image normalization to model

* Remove legacy tensor constructors

Also move away from Int and instead use int64

* Bugfix in MultiscaleRoiAlign

* Move transforms to its own file

* Add missing file

* Lint

* More lint

* Add some basic test for detection models

* More lint

Also move weights from ShuffleNet to PyTorch bucket. Additionally, rename shufflenet to make it consistent with the other models

* Initial version of the segmentation examples

WIP

* Cleanups

* [WIP]

* Tag where runs are being executed

* Minor additions

* Update model with new resnet API

* [WIP] Using torchvision datasets

* Improving datasets

Leverage more and more torchvision datasets

* Reorganizing datasets

* PEP8

* No more SegmentationModel

Also remove outplanes from ResNet, and add a function for querying intermediate outputs. I won't keep it in the end, because it's very hacky and don't work with tracing

* Minor cleanups

* Moving transforms to its own file

* Move models to torchvision

* Bugfixes

* Multiply LR by 10 for classifier

* Remove classifier x 10

* Add tests for segmentation models

* Update with latest utils from classification

* Lint and missing import