GitLab

* Measuring execution times of models.

* Speed up models by avoiding re-estimation of eager output

* Fixing linter

* Reduce input size for big models

* Speed up jit check method.

* Add simple jitscript fallback check for flaky models.

* Restore pytest filtering

* Fixing linter

* 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.

* 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.

* Adding conv_stem support

* fix lint

* bug fix

* address comments

* fix after merge

* adding back checking lines

* fix failing tests

* fix iignore

* add unittest & address comments

* fix memory issue

* address comments

* 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>

* [vit] Adding ViT to torchvision/models

* adding pre-logits layer + resolving comments

* Fix the model attribute bug

* Change version to arch

* fix failing unittests

* remove useless prints

* reduce input size to fix unittests

* Increase windows-cpu executor to 2xlarge

* Use `batch_first=True` and remove classifier

* Change resource_class back to xlarge

* Remove vit_h_14

* Remove vit_h_14 from __all__

* Move vision_transformer.py into prototype

* Fix formatting issue

* remove arch in builder

* Fix type err in model builder

* address comments and trigger unittests

* remove the prototype import in torchvision.models

* Adding vit back to models to trigger CircleCI test

* fix test_jit_forward_backward

* Move all to prototype.

* Adopt new helper methods and fix prototype tests.

* Remove unused import.
Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>
Co-authored-by: Vasilis Vryniotis <vvryniotis@fb.com>

* Refactoring resnet_fpn backbone building.

* Passing the change to *_rcnn and retinanet.

* Applying for faster_rcnn + mobilenetv3

* Applying for ssdlite + mobilenetv3

* Applying for ssd + vgg16

* Update the expected file of retinanet_resnet50_fpn to fix order of initialization.

* Adding full model weights for the VGG16 features.

* adding tests to check output of quantized models

* adding test quantized model weights

* merge test_new_quantized_classification_model with test_quantized_classification_model

* adding skipif removed by mistake

* addressing comments from PR

* removing unused argument

* fixing lint errors

* changing model to eval model and updating weights

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

* enforce single test in circleci

* changing random seed

* updating weights for new seed

* adding missing empty line

* try 128 random seed

* try 256 random seed

* try 16 random seed

* disable inception_v3 input/output quantization tests

* removing ModelTester.test_inception_v3_quantized_expect.pkl

* reverting temporary ci run_test.sh changes
Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>

* 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

* 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.

using nn.init.trunc_normal_ instead of scipy.stats.truncnorm
Co-authored-by: Vincent Moens <vmoens@fb.com>

* Speed up keypoint and retina.

* Update keypoint expected file

* Speed up fasterrcnn_resnet50_fpn.

* Speed up maskrcnn_resnet50_fpn.

* Updating params to resolve flakiness.

* limit runs to those 4 tests

* Relaxing precision to resolve flakiness

* Undo test filtering

* 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.

* Making _segm_resnet() generic and reusable.

* Adding fcn and deeplabv3 directly on mobilenetv3 backbone.

* Adding tests for segmentation models.

* Rename is_strided with _is_cn.

* Add dilation support on MobileNetV3 for Segmentation.

* Add Lite R-ASPP with MobileNetV3 backbone.

* Add pretrained model weights.

* Removing model fcn_mobilenet_v3_large.

* Adding docs and imports.

* Fixing typo and readme.

* Modify segmentation tests compare against expected values.

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

* Tag fasterrcnn mobilenetv3 model with 320, add new inference config that makes it 2x faster sacrificing a bit of mAP.

* Add a high resolution fasterrcnn mobilenetv3 model.

* Update tests and expected values.

* Minor refactoring of a private method to make it reusuable.

* Adding a FasterRCNN + MobileNetV3 with & w/o FPN models.

* Reducing Resolution to 320-640 and anchor sizes to 16-256.

* Increase anchor sizes.

* Adding rpn score threshold param on the train script.

* Adding trainable_backbone_layers param on the train script.

* Adding rpn_score_thresh param directly in fasterrcnn_mobilenet_v3_large_fpn.

* Remove fasterrcnn_mobilenet_v3_large prototype and update expected file.

* Update documentation and adding weights.

* Use buildin Identity.

* Fix spelling.

* Add MobileNetV3 Architecture in TorchVision (#3182)

* Adding implementation of network architecture

* Adding rmsprop support on the train.py

* Adding auto-augment and random-erase in the training scripts.

* Adding support for reduced tail on MobileNetV3.

* Tagging blocks with comments.

* Adding documentation, pre-trained model URL and a minor refactoring.

* Handling better untrained supported models.

* Change children() to modules() to ensure init happens in all blocks.

* Update expected values of all detection models.

* Revert "Update expected values of all detection models."

This reverts commit 050b64ae

* Update expecting values.

* Simplify the ACCEPT=True logic in assertExpected().

* Separate the expected filename estimation from assertExpected

* Unflatten expected values.

* Assert for duplicate scores if primary check fails.

* Remove custom exceptions for algorithms and add a compact function for shrinking large ouputs.

* Removing unused variables.

* Add warning and comments.

* Re-enable all autocast unit-test for detection and marking the tests as skipped in partial validation.

* Move test skip at the end.

* Changing the warning message.

* Vectorize operations, across all feaure levels.

* Remove unnecessary other_outputs variable.

* Split per feature level.

* Perform batched_nms across feature levels.

* Add extra parameter for limiting detections before and after nms.

* Restoring default threshold.

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

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

* Modify expected value and threshold for retinanet unit-test.

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

* Add rough implementation of RetinaNet.

* Move AnchorGenerator to a seperate file.

* Move box similarity to Matcher.

* Expose extra blocks in FPN.

* Expose retinanet in __init__.py.

* Use P6 and P7 in FPN for retinanet.

* Use parameters from retinanet for anchor generation.

* General fixes for retinanet model.

* Implement loss for retinanet heads.

* Output reshaped outputs from retinanet heads.

* Add postprocessing of detections.

* Small fixes.

* Remove unused argument.

* Remove python2 invocation of super.

* Add postprocessing for additional outputs.

* Add missing import of ImageList.

* Remove redundant import.

* Simplify class correction.

* Fix pylint warnings.

* Remove the label adjustment for background class.

* Set default score threshold to 0.05.

* Add weight initialization for regression layer.

* Allow training on images with no annotations.

* Use smooth_l1_loss with beta value.

* Add more typehints for TorchScript conversions.

* Fix linting issues.

* Fix type hints in postprocess_detections.

* Fix type annotations for TorchScript.

* Fix inconsistency with matched_idxs.

* Add retinanet model test.

* Add missing JIT annotations.

* Remove redundant model construction

Make tests pass

* Fix bugs during training on newer PyTorch and unused params in DDP

Needs cleanup and to add back support for images with no annotations

* Cleanup resnet_fpn_backbone

* Use L1 loss for regression

Gives 1mAP improvement over smooth l1

* Disable support for images with no annotations

Need to fix distributed first

* Fix retinanet tests

Need to deduplicate those box checks

* Fix Lint

* Add pretrained model

* Add training info for retinanet
Co-authored-by: Hans Gaiser <hansg91@gmail.com>
Co-authored-by: Hans Gaiser <hans.gaiser@robovalley.com>
Co-authored-by: Hans Gaiser <hans.gaiser@robohouse.com>

* correctness test implemented with old test architecture

* reverted an unneeded change, ran flake8

* moving to relative tolerance of 1 part in 10k for classification correctness checks

* going down to 1 part in 1000 for correctness checks bc architecture differences

* one percent relative tolerance

This reverts commit 1e857d93.

* added correctness tests for classification models
* refactored tests for extensibility & usability
* flake8 fixes

* add expected result tests

* fix wrong assertion

* start with only detection models

* remove unneeded rng setting

* fix test

* add tuple support

* update test

* syntax error

* treat .pkl files as binary data, see : https://git-scm.com/book/en/v2/Customizing-Git-Git-Attributes#_binary_files

* fix test

* fix elif

* Map tensor results and enforce maximum pickle size

* unrelated change

* larger rtol

* pass rtol atol around

* last commit i swear...

* respond to comments

* fix flake

* fix py2 flake