GitLab

* Adding MViT v2 architecture (#6105)

* Adding mvitv2 architecture

* Fixing memory issues on tests and minor refactorings.

* Adding input validation

* Adding docs and minor refactoring

* Add `min_temporal_size` in the supported meta-data.

* Switch Tuple[int, int, int] with List[int] to support easier the 2D case

* Adding more docs and references

* Change naming conventions of classes to follow the same pattern as MobileNetV3

* Fix test breakage.

* Update todos

* Performance optimizations.

* Add support to MViT v1 (#6179)

* Switch implementation to v1 variant.

* Fix docs

* Adding back a v2 pseudovariant

* Changing the way the network are configured.

* Temporarily removing v2

* Adding weights.

* Expand _squeeze/_unsqueeze to support arbitrary dims.

* Update references script.

* Fix tests.

* Fixing frames and preprocessing.

* Fix std/mean values in transforms.

* Add permanent Dropout and update the weights.

* Update accuracies.

* Fix documentation

* Remove unnecessary expected file.

* Skip big model test

* Rewrite the configuration logic to reduce LOC.

* Fix mypy

* Add new .. betastatus:: directive to document Beta APIs

* Also add it for the fine-grained video API

* Add directive for all builders and pages of Detection module

* Also segmentation and video models

* Remove `(N, T, H, W, C) => (N, T, C, H, W)` conversion on presets

* Update docs.

* Fix the tests

* Use `output_format` for `read_video()`

* Use `output_format` for `Kinetics()`

* Adding input descriptions on presets

* Remove models.rst

* Remove '- New'

* Put back torchhub section where it originally was

* Add resnext101_64x4d model definition

* Add test for resnext101_64x4d

* Add resnext101_64x4d weight

* Update checkpoint to use EMA weigth

* Add quantization model signature for resnext101_64x4d

* Fix class name and update accuracy using 1 gpu and batch_size=1

* Apply ufmt

* Update the quantized weight and accuracy that we still keep the training log

* Add quantized expect file

* Update docs and fix acc1

* Add recipe for quantized to PR

* Update models.rst

* Add inception

* Fix googlenet bug

* Fix bug add file

* Update torchvision/models/inception.py
Co-authored-by: Nicolas Hug <contact@nicolas-hug.com>

* Add shufflenetv2 1.5 and 2.0 weights

* Update recipe

* Add to docs

* Use resize_size=232 for eval and update the result

* Add quantized shufflenetv2 large

* Update docs and readme

* Format with ufmt

* Add to hubconf.py

* Update readme for classification reference

* Fix reference classification readme

* Fix typo on readme

* Update reference/classification/readme

* add swin transformer

* Update swin_transformer.py

* Update swin_transformer.py

* fix lint

* fix lint

* refactor code

* add swin_transformer

* Update swin_transformer.py

* fix bug

* refactor code

* fix lint

* update init_weights

* move shift_window into attention

* refactor code

* fix bug

* Update swin_transformer.py

* Update swin_transformer.py

* fix lint

* add patch_merge

* fix bug

* Update swin_transformer.py

* Update swin_transformer.py

* Update swin_transformer.py

* refactor code

* Update swin_transformer.py

* refactor code

* fix lint

* refactor code

* add swin_tiny

* add swin_tiny.pkl

* fix lint

* Delete ModelTester.test_swin_tiny_expect.pkl

* add swin_tiny

* add

* add Optional to bias

* update init weights

* update init_weights and add no weight decay

* add no weight decay

* add set_weight_decay

* add set_weight_decay

* fix lint

* fix lint

* add lr_cos_min

* add other swin models

* Update torchvision/models/swin_transformer.py
Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>

* refactor doc

* Update utils.py

* Update train.py

* Update train.py

* Update swin_transformer.py

* update model builder

* fix lint

* add

* Update torchvision/models/swin_transformer.py
Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>

* Update torchvision/models/swin_transformer.py
Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>

* update other model

* simplify the model name just like ViT

* add lr_cos_min

* fix lint

* fix lint

* Update swin_transformer.py

* Update swin_transformer.py

* Update swin_transformer.py

* Delete ModelTester.test_swin_tiny_expect.pkl

* add swin_t

* refactor code

* Update train.py

* add swin_s

* ignore a error of mypy

* Update swin_transformer.py

* fix lint

* add swin_b

* add swin_l

* refactor code

* Update train.py

* move relative_position_bias to __init__

* fix formatting

* Revert "fix formatting"

This reverts commit 41faba232668f7ac4273a0cf632c0d0130c7ce9c.

* Revert "move relative_position_bias to __init__"

This reverts commit f0615440bf18617dc0e5dc4839bd5ed27e5ed010.

* refactor code

* Remove deprecated meta-data from `_COMMON_META`

* fix linter

* add pretrained weights for swin_t

* fix format

* apply ufmt

* add documentation

* update references README

* adding new style docs

* update pre-trained weights values

* remove other variants

* fix typo

* Remove expect for the variants not yet supported
Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>
Co-authored-by: Joao Gomes <jdsgomes@fb.com>

* Add vit_b_16_swag

* Better handling idiom for image_size, edit test_extended_model to handle case where number of param differ from default due to different image size input

* Update the accuracy to the experiment result on torchvision model

* Fix typo missing underscore

* raise exception instead of torch._assert, add back publication year (accidentally deleted)

* Add license information on meta and readme

* Improve wording and fix typo for pretrained model license in readme

* Add vit_l_16 weight

* Update README.rst
Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>

* Update the accuracy meta on vit_l_16_swag model to result from our experiment

* Add vit_h_14_swag model

* Add accuracy from experiments

* Add to vit_h_16 model to hubconf.py

* Add docs and expected pkl file for test

* Remove legacy compatibility for ViT_H_14 model
Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>

* Test vit_h_14 with smaller image_size to speedup the test
Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>

Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>

* Moving basefiles outside of prototype and porting Alexnet, ConvNext, Densenet and EfficientNet.

* Porting googlenet

* Porting inception

* Porting mnasnet

* Porting mobilenetv2

* Porting mobilenetv3

* Porting regnet

* Porting resnet

* Porting shufflenetv2

* Porting squeezenet

* Porting vgg

* Porting vit

* Fix docstrings

* Fixing imports

* Adding missing import

* Fix mobilenet imports

* Fix tests

* Fix prototype tests

* Exclude get_weight from models on test

* Fix init files

* Porting googlenet

* Porting inception

* porting mobilenetv2

* porting mobilenetv3

* porting resnet

* porting shufflenetv2

* Fix test and linter

* Fixing docs.

* Porting Detection models (#5617)

* fix inits

* fix docs

* Port faster_rcnn

* Port fcos

* Port keypoint_rcnn

* Port mask_rcnn

* Port retinanet

* Port ssd

* Port ssdlite

* Fix linter

* Fixing tests

* Fixing tests

* Fixing vgg test

* Porting Optical Flow, Segmentation, Video models (#5619)

* Porting raft

* Porting video resnet

* Porting deeplabv3

* Porting fcn and lraspp

* Fixing the tests and linter

* Porting docs, examples, tutorials and galleries (#5620)

* Fix examples, tutorials and gallery

* Update gallery/plot_optical_flow.py
Co-authored-by: Nicolas Hug <contact@nicolas-hug.com>

* Fix import

* Revert hardcoded normalization

* fix uncommitted changes

* Fix bug

* Fix more bugs

* Making resize optional for segmentation

* Fixing preset

* Fix mypy

* Fixing documentation strings

* Fix flake8

* minor refactoring
Co-authored-by: Nicolas Hug <contact@nicolas-hug.com>

* Resolve conflict

* Porting model tests (#5622)

* Porting tests

* Remove unnecessary variable

* Fix linter

* Move prototype to extended tests

* Fix download models job

* Update CI on Multiweight branch to use the new weight download approach (#5628)

* port Pad to prototype transforms (#5621)

* port Pad to prototype transforms

* use literal

* Bump up LibTorchvision version number for Podspec to release Cocoapods (#5624)
Co-authored-by: Anton Thomma <anton@pri.co.nz>
Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>

* pre-download model weights in CI docs build (#5625)

* pre-download model weights in CI docs build

* move changes into template

* change docs image

* Regenerated config.yml
Co-authored-by: Philip Meier <github.pmeier@posteo.de>
Co-authored-by: Anton Thomma <11010310+thommaa@users.noreply.github.com>
Co-authored-by: Anton Thomma <anton@pri.co.nz>

* Porting reference scripts and updating presets (#5629)

* Making _preset.py classes

* Remove support of targets on presets.

* Rewriting the video preset

* Adding tests to check that the bundled transforms are JIT scriptable

* Rename all presets from *Eval to *Inference

* Minor refactoring

* Remove --prototype and --pretrained from reference scripts

* remove  pretained_backbone refs

* Corrections and simplifications

* Fixing bug

* Fixing linter

* Fix flake8

* restore documentation example

* minor fixes

* fix optical flow missing param

* Fixing commands

* Adding weights_backbone support in detection and segmentation

* Updating the commands for InceptionV3

* Setting `weights_backbone` to its fully BC value (#5653)

* Replace default `weights_backbone=None` with its BC values.

* Fixing tests

* Fix linter

* Update docs.

* Update preprocessing on reference scripts.

* Change qat/ptq to their full values.

* Refactoring preprocessing

* Fix video preset

* No initialization on VGG if pretrained

* Fix warning messages for backbone utils.

* Adding star to all preset constructors.

* Fix mypy.
Co-authored-by: Nicolas Hug <contact@nicolas-hug.com>
Co-authored-by: Philip Meier <github.pmeier@posteo.de>
Co-authored-by: Anton Thomma <11010310+thommaa@users.noreply.github.com>
Co-authored-by: Anton Thomma <anton@pri.co.nz>

* Remove from models and references.

* Adding most tests and docs.

* Adding transforms tests.

* Remove unnecesary ipython notebook.

* Simplify tests.

* Addressing comments.

* Extend the EfficientNet class to support v1 and v2.

* Refactor config/builder methods and add prototype builders

* Refactoring weight info.

* Update dropouts based on TF config ref

* Update BN eps on TF base_config

* Use Conv2dNormActivation.

* Adding pre-trained weights for EfficientNetV2-s

* Add Medium and Large weights

* Update stats with single batch run.

* Add accuracies in the docs.

* Graduate ConvNeXt to main TorchVision area.

* Linter and all var.

* Renaming var and making named params mandatory.

* Refactor model builder

* Add 3 more convnext variants.

* Adding weights for convnext_small.

* Fix minor bug.

* Fix number of parameters for small model.

* Adding weights for the base variant.

* Adding weights for the large variant.

* Simplify LayerNorm2d implementation.

* Optimize speed of CNBlock.

* Repackage weights.

* add fcos

* update fcos

* add giou_loss

* add BoxLinearCoder for FCOS

* add full code for FCOS

* add giou loss

* add fcos

* add __all__

* Fixing lint

* Fixing lint in giou_loss.py

* Add typing annotation to fcos

* Add trained checkpoints

* Use partial to replace lambda

* Minor fixes to docstrings

* Apply ufmt format

* Fixing docstrings

* Fixing jit scripting

* Minor fixes to docstrings

* Fixing jit scripting

* Ignore mypy in fcos

* Fixing trained checkpoints

* Fixing unit-test of jit script

* Fixing docstrings

* Add test/expect/ModelTester.test_fcos_resnet50_fpn_expect.pkl

* Fixing test_detection_model_trainable_backbone_layers

* Update test_fcos_resnet50_fpn_expect.pkl

* rename stride to box size

* remove TODO and fix some typo

* merge some code for better

* impove the comments

* remove decode and encode of BoxLinearCoder

* remove some unnecessary hints

* use the default value in detectron2.

* update doc

* Add unittest for BoxLinearCoder

* Add types in FCOS

* Add docstring for BoxLinearCoder

* Minor fix for the docstring

* update doc

* Update fcos_resnet50_fpn_coco pretained weights url

* Update torchvision/models/detection/fcos.py
Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>

* Update torchvision/models/detection/fcos.py
Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>

* Update torchvision/models/detection/fcos.py
Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>

* Update torchvision/models/detection/fcos.py
Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>

* Add FCOS model documentation

* Fix typo in FCOS documentation

* Add fcos to the prototype builder

* Capitalize COCO_V1

* Fix params of fcos

* fix bug for partial

* Fixing docs indentation

* Fixing docs format in giou_loss

* Adopt Reference for GIoU Loss

* Rename giou_loss to generalized_box_iou_loss

* remove overwrite_eps

* Update AP test values

* Minor fixes for the docs

* Minor fixes for the docs

* Update torchvision/models/detection/fcos.py
Co-authored-by: Zhiqiang Wang <zhiqwang@foxmail.com>

* Update torchvision/prototype/models/detection/fcos.py
Co-authored-by: Zhiqiang Wang <zhiqwang@foxmail.com>
Co-authored-by: zhiqiang <zhiqwang@foxmail.com>
Co-authored-by: Joao Gomes <jdsgomes@fb.com>
Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>
Co-authored-by: Joao Gomes <joaopsgomes@gmail.com>

* Adding CNBlock and skeleton architecture

* Completed implementation.

* Adding model in prototypes.

* Add test and minor refactor for JIT.

* Fix mypy.

* Fixing naming conventions.

* Fixing tests.

* Fix stochastic depth percentages.

* Adding stochastic depth to tiny variant.

* Minor refactoring and adding comments.

* Adding weights.

* Update default weights.

* Fix transforms issue

* Move convnext to prototype.

* linter fix

* fix docs

* Addressing code review comments.

* adding vit_h_14

* prototype and docs

* bug fix

* adding curl check

* add regnet_y_128gf

* fix test

* add expected test file

* update regnet factory function, add to prototype as well

* write torchscript to temp file instead bytesio in model test

* docs

* clear GPU memory

* no_grad

* nit
Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>

* graduate vit from prototype

* nit

* add vit to docs and hubconf

* ufmt

* re-correct ufmt

* again

* fix linter

Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>

* Clean up unnecessary quant builders and add quant weights for 0.5

* Fixing mypy.

* Fix TORCH_HOME environment variable in docs

* Update docs with torch.hub

Updated the reference for loading a model to `torch.hub`.
Co-authored-by: Nicolas Hug <contact@nicolas-hug.com>
Co-authored-by: Nicolas Hug <contact@nicolas-hug.com>

* add best weights and x_1_6, x_3_2, y_1_6, y_3_2, y_32 weights

* add best weights and x_1_6, x_3_2, y_1_6, y_3_2, y_32 weights

* add weights for x_16gf, x_32gf, y_16gf

* initial code

* add SqueezeExcitation

* initial code

* add SqueezeExcitation

* add SqueezeExcitation

* regnet blocks, stems and model definition

* nit

* add fc layer

* use Callable instead of Enum for block, stem and activation

* add regnet_x and regnet_y model build functions, add docs

* remove unused depth

* use BN/activation constructor and ConvBNActivation

* add expected test pkl files

* allow custom activation in SqueezeExcitation

* use ReLU as the default activation

* initial code

* add SqueezeExcitation

* initial code

* add SqueezeExcitation

* add SqueezeExcitation

* regnet blocks, stems and model definition

* nit

* add fc layer

* use Callable instead of Enum for block, stem and activation

* add regnet_x and regnet_y model build functions, add docs

* remove unused depth

* use BN/activation constructor and ConvBNActivation

* reuse SqueezeExcitation from efficientnet

* refactor RegNetParams into BlockParams

* use nn.init, replace np with torch

* update README

* construct model with stem, block, classifier instances

* Revert "construct model with stem, block, classifier instances"

This reverts commit 850f5f3ed01a2a9b36fcbf8405afd6e41d2e58ef.

* remove unused blocks

* support scaled model

* fuse into ConvBNActivation

* make reset_parameters private

* fix type errors

* fix for unit test

* add pretrained weights for 6 variant models, update docs

Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>

* Adding code skeleton

* Adding MBConvConfig.

* Extend SqueezeExcitation to support custom min_value and activation.

* Implement MBConv.

* Replace stochastic_depth with operator.

* Adding the rest of the EfficientNet implementation

* Update torchvision/models/efficientnet.py

* Replacing 1st activation of SE with SiLU.

* Adding efficientnet_b3.

* Replace mobilenetv3 assets with custom.

* Switch to standard sigmoid and reconfiguring BN.

* Reconfiguration of efficientnet.

* Add repr

* Add weights.

* Update weights.

* Adding B5-B7 weights.

* Update docs and hubconf.

* Fix doc link.

* Fix typo on comment.

* Partial implementation of SSDlite.

* Add normal init and BN hyperparams.

* Refactor to keep JIT happy

* Completed SSDlite.

* Fix lint

* Update todos

* Add expected file in repo.

* Use C4 expansion instead of C4 output.

* Change scales formula for Default Boxes.

* Add cosine annealing on trainer.

* Make T_max count epochs.

* Fix test and handle corner-case.

* Add support of support width_mult

* Add ssdlite presets.

* Change ReLU6, [-1,1] rescaling, backbone init & no pretraining.

* Use _reduced_tail=True.

* Add sync BN support.

* Adding the best config along with its weights and documentation.

* Make mean/std configurable.

* Fix not implemented for half exception

* Early skeleton of API.

* Adding MultiFeatureMap and vgg16 backbone.

* Making vgg16 backbone same as paper.

* Making code generic to support all vggs.

* Moving vgg's extra layers a separate class + L2 scaling.

* Adding header vgg layers.

* Fix maxpool patching.

* Refactoring code to allow for support of different backbones & sizes:
- Skeleton for Default Boxes generator class
- Dynamic estimation of configuration when possible
- Addition of types

* Complete the implementation of DefaultBox generator.

* Replace randn with empty.

* Minor refactoring

* Making clamping between 0 and 1 optional.

* Change xywh to xyxy encoding.

* Adding parameters and reusing objects in constructor.

* Temporarily inherit from Retina to avoid dup code.

* Implement forward methods + temp workarounds to inherit from retina.

* Inherit more methods from retinanet.

* Fix type error.

* Add Regression loss.

* Fixing JIT issues.

* Change JIT workaround to minimize new code.

* Fixing initialization bug.

* Add classification loss.

* Update todos.

* Add weight loading support.

* Support SSD512.

* Change kernel_size to get output size 1x1

* Add xavier init and refactoring.

* Adding unit-tests and fixing JIT issues.

* Add a test for dbox generator.

* Remove unnecessary import.

* Workaround on GeneralizedRCNNTransform to support fixed size input.

* Remove unnecessary random calls from the test.

* Remove more rand calls from the test.

* change mapping and handling of empty labels

* Fix JIT warnings.

* Speed up loss.

* Convert 0-1 dboxes to original size.

* Fix warning.

* Fix tests.

* Update comments.

* Fixing minor bugs.

* Introduce a custom DBoxMatcher.

* Minor refactoring

* Move extra layer definition inside feature extractor.

* handle no bias on init.

* Remove fixed image size limitation

* Change initialization values for bias of classification head.

* Refactoring and update test file.

* Adding ResNet backbone.

* Minor refactoring.

* Remove inheritance of retina and general refactoring.

* SSD should fix the input size.

* Fixing messages and comments.

* Silently ignoring exception if test-only.

* Update comments.

* Update regression loss.

* Restore Xavier init everywhere, update the negative sampling method, change the clipping approach.

* Fixing tests.

* Refactor to move the losses from the Head to the SSD.

* Removing resnet50 ssd version.

* Adding support for best performing backbone and its config.

* Refactor and clean up the API.

* Fix lint

* Update todos and comments.

* Adding RandomHorizontalFlip and RandomIoUCrop transforms.

* Adding necessary checks to our tranforms.

* Adding RandomZoomOut.

* Adding RandomPhotometricDistort.

* Moving Detection transforms to references.

* Update presets

* fix lint

* leave compose and object

* Adding scaling for completeness.

* Adding params in the repr

* Remove unnecessary import.

* minor refactoring

* Remove unnecessary call.

* Give better names to DBox* classes

* Port num_anchors estimation in generator

* Remove rescaling and fix presets

* Add the ability to pass a custom head and refactoring.

* fix lint

* Fix unit-test

* Update todos.

* Change mean values.

* Change the default parameter of SSD to train the full VGG16 and remove the catch of exception for eval only.

* Adding documentation

* Adding weights and updating readmes.

* Update the model weights with a more performing model.

* Adding doc for head.

* Restore import.

* Adding TODO placeholders.

* More placeholders.

* Add MobileNetV3 small pre-trained weights.

* Remove placeholders.