Skip to content

Releases: Lightning-Universe/lightning-bolts

Improved YOLO models

30 Jun 11:38
Compare
Choose a tag to compare

[0.7.0] - 2022-06-30

Added

  • Improved YOLO model includes YOLOv4, YOLOv5, and YOLOX networks and training algorithms (#817)

Changed

  • Move SSL transforms to pl_bolts/transforms (#905)
  • Reviewed models.detection.yolo (#851)
  • Reviewed LogisticRegression (#950)
  • Bumped support of min python version to py3.8+ (#1021)
  • Update numpy compatibility to <1.25.0 (#959)
  • Update torchmetrics compatibility to <0.12.0 (#1016)
  • Update pytorch-lightning compatibility to >1.7.0,<2.0.0 (#965, #973, #1006)

Fixed

  • Dropped reference to torch._six (#993)

New Contributors

Full Changelog: 0.6.0...0.7.0

Minor patch release

16 Dec 13:31
Compare
Choose a tag to compare

What's Changed

Full Changelog: 0.6.0...0.6.0.post1

Revisions and compatibility

03 Nov 11:01
06a4f9f
Compare
Choose a tag to compare

[0.6.0] - 2022-11-03

Added

  • Updated SparseML callback for latest PyTorch Lightning (#822)

  • Updated torch version to v1.10.X (#815)

  • Dataset specific args method to CIFAR10, ImageNet, MNIST, and STL10 (#890)

  • Migrate to use lightning-utilities (#907)

  • Support PyTorch Lightning v1.8 (#910)

  • Major revision of Bolts

    • under_review flag (#835, #837)
    • Reviewing GAN basics, VisionDataModule, MNISTDataModule, CIFAR10DataModule (#843)
    • Added tests, updated doc-strings for Dummy Datasets (#865)
    • Binary MNIST/EMNIST Datasets and Datamodules (#866)
    • FashionMNIST/EMNIST Datamodules (#871)
    • Revision ArrayDataset (#872)
    • BYOL weight update callback (#867)
    • Revision models.vision.unet, models.vision.segmentation (#880)
    • Revision of SimCLR transforms (#857)
    • Revision Metrics (#878, #887)
    • Revision of BYOL module and tests (#874)
    • Revision of MNIST module (#873)
    • Revision of dataset normalizations (#898)
    • Revision of SimSiam module and tests (#891)
    • Revision datasets.kitti_dataset.KittiDataset (#896)
    • SWAV improvements (#903)
    • minor dcgan-import fix (#921)

Fixed

Contributors

@ArnolFokam, @Atharva-Phatak, @BaruchG, @Benjamin-Etheredge, @Borda, @Ce11an, @clementpoiret, @kfirgedal, @lijm1358, @matsumotosan, @nishantb06, @otaj, @rohitgr7, @shivammehta25, @TrellixVulnTeam

If we forgot someone due to not matching commit email with GitHub account, let us know :]

More object detection models and backbones

20 Dec 17:15
ba899e8
Compare
Choose a tag to compare

[0.5.0] - 2021-12-20

Added

  • Added YOLO model (#552)
  • Added SRGAN, SRImageLoggerCallback, TVTDataModule, SRCelebA, SRMNIST, SRSTL10 (#466)
  • Added nn.Module support for FasterRCNN backbone (#661)
  • Added RetinaNet with torchvision backbones (#529)
  • Added Python 3.9 support (#786)

Changed

  • VAE now uses deterministic KL divergence during training, previously estimated KL divergence by random sampling (#760)

Removed

  • Removed PyTorch 1.6 support (#786)
  • Removed Python 3.6 support (#785)

Fixed

  • Fixed doctest fails with ImportError: cannot import name 'Env' from 'gym' (#751)
  • Fixed MoCo v2 missing Cosine Annealing learning scheduler (#757)

Contributors

@abhayraw1 @akihironitta @chris-clem @hoangtnm @nmichlo @oke-aditya @Programmer-RD-AI @senarvi

If we forgot someone due to not matching commit email with GitHub account, let us know :]

More RL and callbacks

09 Sep 19:25
Compare
Choose a tag to compare

[0.4.0] - 2021-09-09

Added

  • Added Soft Actor Critic (SAC) Model (#627)
  • Added EMNISTDataModule, BinaryEMNISTDataModule, and BinaryEMNIST dataset (#676)
  • Added Advantage Actor-Critic (A2C) Model (#598)
  • Added Torch ORT Callback (#720)
  • Added SparseML Callback (#724)

Changed

  • Changed the default values pin_memory=False, shuffle=False and num_workers=16 to pin_memory=True, shuffle=True and num_workers=0 of datamodules (#701)
  • Supporting deprecated attribute usage (#699)

Fixed

  • Fixed ImageNet val loader to use val transform instead of train transform (#713)
  • Fixed the MNIST download giving HTTP 404 with torchvision>=0.9.1 (#674)
  • Removed momentum updating from val step and add separate val queue (#631)
  • Fixed moving the queue to GPU when resuming checkpoint for SwAV model (#684)
  • Fixed FP16 support with vision GPT model (#694)
  • Removing bias from linear model regularisation (#669)
  • Fixed CPC module issue (#680)

less softmax

17 Jun 16:01
Compare
Choose a tag to compare

[0.3.4] - 2021-06-17

Changed

  • Replaced load_boston with load_diabetes in the docs and tests (#629)
  • Added base encoder and MLP dimension arguments to BYOL constructor (#637)

Fixed

  • Fixed the MNIST download giving HTTP 503 (#633)
  • Fixed type annotation of ExperienceSource.__iter__ (#645)
  • Fixed pretrained_urls on Windows (#652)
  • Fixed logistic regression (#655, #664)
  • Fixed double softmax in SSLEvaluator (#663)

updated LARS

17 Apr 14:09
Compare
Choose a tag to compare

[0.3.3] - 2021-04-17

Changed

  • Suppressed missing package warnings, conditioned by WARN_MISSING_PACKAGE="1" (#617)
  • Updated all scripts to LARS (#613)

Fixed

  • Add missing dataclass requirements (#618)

typing friendly

20 Mar 00:43
Compare
Choose a tag to compare

[0.3.2] - 2021-03-20

Changed

  • Renamed SSL modules: CPCV2 >> CPC_v2 and MocoV2 >> Moco_v2 (#585)
  • Refactored setup.py to be typing friendly (#601)

compatibility PyTorch 1.8

09 Mar 18:13
Compare
Choose a tag to compare

[0.3.1] - 2021-03-09

Added

  • Added Pix2Pix model (#533)

Changed

  • Moved vision models (GPT2, ImageGPT, SemSegment, UNet) to pl_bolts.models.vision (#561)

Fixed

  • Fixed BYOL moving average update (#574)
  • Fixed custom gamma in rl (#550)
  • Fixed PyTorch 1.8 compatibility issue (#580, #579)
  • Fixed handling batchnorms in BatchGradientVerification [#569)
  • Corrected num_rows calculation in LatentDimInterpolator callback (#573)

Contributors

@akihironitta, @aniketmaurya, @BartekRoszak, @FlorianMF, @indigoviolet, @kaushikb11, @mxksowie, @wjn0

If we forgot someone due to not matching commit email with GitHub account, let us know :]

major fixes & refactoring

20 Jan 23:24
Compare
Choose a tag to compare

Detail chnages

Added

  • Added input_channels argument to UNet (#297)
  • Added SwAV (#239, #348, #323)
  • Added data monitor callbacks ModuleDataMonitor and TrainingDataMonitor (#285)
  • Added DCGAN module (#403)
  • Added VisionDataModule as parent class for BinaryMNISTDataModule, CIFAR10DataModule, FashionMNISTDataModule,
    and MNISTDataModule (#400)
  • Added GIoU loss (#347)
  • Added IoU loss (#469)
  • Added semantic segmentation model SemSegment with UNet backend (#259)
  • Added option to normalize latent interpolation images (#438)
  • Added flags to datamodules (#388)
  • Added metric GIoU (#347)
  • Added Intersection over Union Metric/Loss (#469)
  • Added SimSiam model (#407)
  • Added gradient verification callback (#465)
  • Added Backbones to FRCNN (#475)

Changed

  • Decoupled datamodules from models (#332, #270)
  • Set PyTorch Lightning 1.0 as the minimum requirement (#274)
  • Moved pl_bolts.callbacks.self_supervised.BYOLMAWeightUpdate to pl_bolts.callbacks.byol_updates.BYOLMAWeightUpdate (#288)
  • Moved pl_bolts.callbacks.self_supervised.SSLOnlineEvaluator to pl_bolts.callbacks.ssl_online.SSLOnlineEvaluator (#288)
  • Moved pl_bolts.datamodules.*_dataset to pl_bolts.datasets.*_dataset (#275)
  • Ensured sync across val/test step when using DDP (#371)
  • Refactored CLI arguments of models (#394)
  • Upgraded DQN to use .log (#404)
  • Decoupled DataModules from models - CPCV2 (#386)
  • Refactored datamodules/datasets (#338)
  • Refactored Vision DataModules (#400)
  • Refactored pl_bolts.callbacks (#477)
  • Refactored the rest of pl_bolts.models.self_supervised (#481, #479)
  • Update [torchvision.utils.make_grid(https://pytorch.org/docs/stable/torchvision/utils.html#torchvision.utils.make_grid)] kwargs to TensorboardGenerativeModelImageSampler (#494)

Fixed

  • Fixed duplicate warnings when optional packages are unavailable (#341)
  • Fixed ModuleNotFoundError when importing datamoules (#303)
  • Fixed cyclic imports in pl_bolts.utils.self_suprvised (#350)
  • Fixed VAE loss to use KL term of ELBO (#330)
  • Fixed dataloders of MNISTDataModule to use self.batch_size (#331)
  • Fixed missing outputs in SSL hooks for PyTorch Lightning 1.0 (#277)
  • Fixed stl10 datamodule (#369)
  • Fixes SimCLR transforms (#329)
  • Fixed binary MNIST datamodule (#377)
  • Fixed the end of batch size mismatch (#389)
  • Fixed batch_size parameter for DataModules remaining (#344)
  • Fixed CIFAR num_samples (#432)
  • Fixed DQN run_n_episodes using the wrong environment variable (#525)

Contributors

@akihironitta, @ananyahjha93, @annikabrundyn, @awaelchli, @Borda, @briankosw, @chris-clem, @deng-cy, @hecoding, @miccio-dk, @oke-aditya, @SeanNaren, @sid-sundrani, @teddykoker, @zlapp

If we forgot someone due to not matching commit email with GitHub account, let us know :]