GitLab

* Enhance ShufflenetV2

Class shufflenetv2 receives `stages_repeats` and `stages_out_channels` arguments.

* remove explicit num_classes argument from utility functions

* Minor refactoring of ShuffleNetV2

Added progress flag following #875. Further the following refactoring was also done:

1) added `version` argument in shufflenetv2 method and removed the operations for converting the `width_mult` arg to float and string.
2) removed `num_classes` argument and **kwargs from functions except `ShuffleNetV2`

* removed `version` arg

* Update shufflenetv2.py

* Removed the try except block

* Update shufflenetv2.py

* Changed version from float to str

* Replace `width_mult` with `stages_out_channels`

Removes the need of  `_getStages` function.

* remove 'input_size' parameter from shufflenetv2

* Update shufflenetv2.py

* Add ShuffleNet v2

Added 4 configurations: x0.5, x1, x1.5, x2
Add 2 pretrained models: x0.5, x1

* fix lint

* Change globalpool to torch.mean() call

* added progress flag to model getters

* flake8

* bug fix

* backward commpability

* Add dilation option to ResNet

* Add a size check for replace_stride_with_dilation

* Fix ResNeXt model defs with backwards compat for ResNet.

* Fix Python 2.x integer div issue

* add aux_logits support to inception

it is related to pytorch/pytorch#18668

* instantiate InceptionAux only when requested

it is related to pytorch/pytorch#18668

* revert googlenet

* support and aux_logits in pretrained models

* return namedtuple when aux_logit is True

* Add groups support to ResNet

* Kill BaseResNet

* Make it support multi-machine training

Making references/classification/train.py and references/classification/utils.py  compatible with python2  (#831)

* linter fixes

* linter fixes

* remove duplicate code from densenet

* correct indentation

* Match Tensorflows implementation of GoogLeNet

* just disable the branch when pretrained is true

* don't use legacy code

* Add MobileNet V2

* Remove redundant functions and make tests pass

* Simplify a bit the implementation

* Reuse ConvBNReLU more often

* Remove input_size and minor changes

* Py2 fix

In reference to #729 added comments to clarify the naming and action of the layers performing downsampling in resnets.

* Added dimensions in the comments

The update provides the dimensions of the processed data following the style of inceptionV3 implementation.

* Changed docs and comments

Updated doc with the argument `transform_input`. Modified comments to match inceptionV3 style.

Including the `transform_input` argument in the docs of inceptionV3

* Add GoogLeNet (Inception v1)

* Fix missing padding

* Add missing ReLu to aux classifier

* Add Batch normalized version of GoogLeNet

* Use ceil_mode instead of padding and initialize weights using "xavier"

* Match BVLC GoogLeNet zero initialization of classifier

* Small cleanup

* use adaptive avg pool

* adjust network to match TensorFlow

* Update url of pre-trained model and add classification results on ImageNet

* Bugfix that improves performance by 1 point

* flake_fixes

* flake_fixes2

* Modifying the comments of inceptionV3 dimensions

Modifying the comments of inceptionV3 dimensions to match the pytorch convention. Relevant (https://github.com/pytorch/vision/pull/719#pullrequestreview-203194302)

* Added Batch size in comment

* Update inception.py

* Updated inceptionV3 to accept different sized images (Adaptive avg pool)

The update allows inceptionV3 to process images larger or smaller than prescribed image size (299x299) using adaptive average pooling. Will be useful while finetuning or testing on different resolution images.

* Update inception.py

The update allows VGG to process images larger or smaller than prescribed imagenet size using adaptive average pooling. Will be useful while finetuning or testing on different resolution images. Similar to https://github.com/pytorch/vision/pull/643 and https://github.com/pytorch/vision/pull/672. I did not include adaptive avg pool in features or classifier block so that these predefined blocks can be used as it is.

The update allows Alexnet to process images larger or smaller than prescribed imagenet size using adaptive average pooling. Will be useful while finetuning or testing on different resolution images. Similar to https://github.com/pytorch/vision/pull/643 and https://github.com/pytorch/vision/pull/672. I did not include adaptive avg pool in features or classifier block so that these predefined blocks can be used as it is.

* Added size information for inception v3

Since inception v3 expects in contrast to the other models a size
of 299x299x3 I added a small note in the docstring to point that out
(addresses #560)

* inception size info in note box; fixed line length

* removed trailing whitespace

* Zero-init the residual branch in resnet

* Add zero_init_residual as an option

* Changing to AdaptiveAvgPool2d

* add whitespace

* Update densenet.py

* Update densenet.py

* use predefined function

* cleanup resnet

* fix stride

* revert inplanes since it isnt constant

* lint

* AdaptiveAvgpool2d integrated

The Average Pool Layer on SqueezeNet and ResNet was hardcoded, this was changed, now any the input size is accepted.

* error fixed

* linting fixed

* Fix inception v3 input transform for trace & onnx

* Input transform are in-place updates, which produce issues for tracing
and exporting to onnx.

* nit