What's Changed

Speed up function _same_mask in DetConBLoss by 25% by @misrasaurabh1 in https://github.com/lightly-ai/lightly/pull/1847
Fix DINOv2 Bugs
Add DINOv2 to Docs with Example
Fix: vicregl_loss tensor RuntimeError by @RDR2Blackwater in https://github.com/lightly-ai/lightly/pull/1850
Add iBOTTransform
Add iBOT Implementation
Add iBOT to Docs with Example
Fix KNN dtype for ResNet benchmark
Fix CUDA Tests for Losses
Fix OnlineClassifier Issues
Support Tensor Input for GaussianBlur
Use ToTensor Helper Function by @ajtritt in https://github.com/lightly-ai/lightly/pull/1862
Move Rotation After Flip for Transforms by @KylevdLangemheen in https://github.com/lightly-ai/lightly/pull/1865
Improve pre-commit hooks for Dev Env
Add iBOT Benchmarks

New Contributors

@misrasaurabh1 made their first contribution in https://github.com/lightly-ai/lightly/pull/1847
@RDR2Blackwater made their first contribution in https://github.com/lightly-ai/lightly/pull/1850
@ajtritt made their first contribution in https://github.com/lightly-ai/lightly/pull/1862
@KylevdLangemheen made their first contribution in https://github.com/lightly-ai/lightly/pull/1865

Full Changelog: https://github.com/lightly-ai/lightly/compare/v1.15.21...v1.15.22

Many thanks to our contributors!

Models

- Python
Published by yutong-xiang-97 7 months ago

What's Changed

Add DINOv2 ViT benchmark Implementation
Add Paper Joint-Embedding vs Reconstruction: Provable Benefits of Latent Space Prediction for Self-Supervised Learning, 2025 by Meta to "Lightly in Research". Thank them for the credit!
Add seed_everything for reproducibility in benchmarks by @yvesyue in https://github.com/lightly-ai/lightly/pull/1819
Fix MyPy type-checking issues for newer versions of NumPy by @yvesyue in https://github.com/lightly-ai/lightly/pull/1820
Fix DCLLoss negative-term aggregation and add loop-based reference test by @yvesyue in https://github.com/lightly-ai/lightly/pull/1827
Fix bugs in KNN benchmark evaluation
Fix bugs in cosine scheduler warmup epochs
Fix MaskedCausalBlock.__init__() got an unexpected keyword argument 'proj_bias' due to interface change in the newer TIMM versions
Fix AddGridTransform due to interface change in the newer Torchvision versions
Fix format & format-check to only target python directories
Remove video download functions
Remove unused download functions & add typing

New Contributors

@yvesyue made their first contribution in https://github.com/lightly-ai/lightly/pull/1819

Full Changelog: https://github.com/lightly-ai/lightly/compare/v1.5.20...v1.15.21

Many thanks to our contributors!

Models

- Python
Published by yutong-xiang-97 9 months ago

Changes

Added DINO ViT Benchmark
Fixed BTLoss: ensure invariance to affine transformations by @adosar in https://github.com/lightly-ai/lightly/pull/1806
Tested BTLoss: use default values for torch.allclose by @adosar in https://github.com/lightly-ai/lightly/pull/1810
Added more detailed docstring to knn predict by @maxprogrammer007 in https://github.com/lightly-ai/lightly/pull/1812
Removed verbose in CosineWarmUpScheduler
Add LightlyTrain Reference
Renamed lightly-train command to lightly-ssl-train

Many thanks to our contributors!

Models

- Python
Published by yutong-xiang-97 10 months ago

Updates to the Model Documentations and addition of MACL Loss

Changes

Docs: Fix mismatch between forward method and docstring of NTXentLoss by @adosar in https://github.com/lightly-ai/lightly/pull/1789
Update NNCLR model examples in docs
Updated the BYOL model examples in docs
Updated the DINO model examples in docs
Update the SimSiam model examples in docs
Additional tests added to pooling operation for DetCon
Fix issues with the Lightning Trainer's strategy in the MAE examples and support new Lightning versions in the benchmarks
Added loss function for MACL (Model-Aware Contrastive Learning)
Updated CONTRIBUTING Guide and GitHub Actions
Fix multiple issues with loss tests

Models

- Python
Published by liopeer about 1 year ago

lightly - update SimCLR and MAE model docs

What's Changed

Update SimCLR model model example
Update MAE model docs
adjust pagination to be defensive at 2500 entries
Fix typo: Dict() instead of dict() return class instanciation error by @gatienc in https://github.com/lightly-ai/lightly/pull/1785

Many thanks to all of our contributors!

Models

- Python
Published by japrescott about 1 year ago

lightly - cleanup & detcon loss fix

Changes

fix detcon loss distributed issue
remove heuristics in masked pooling
fix torchvision dependency test

Many thanks to all of our contributors!

Models

- Python
Published by japrescott about 1 year ago

Changes

Added DetConSLoss and DetConBLoss
removed opencv dependencies partial thanks to @vectorvp
fix: IJEPA example by thanks to @vectorvp

Typing and Docs

Many files are now properly typed and type-checked thanks to @philippmwirth
We removed old and outdated documentation regarding the LightlyOne Worker

Many thanks to all of our contributors!

Models

- Python
Published by japrescott about 1 year ago

lightly - More and better Transforms

Changes

New transforms

Add PhaseShift Transform (#1714) by @pearguacamole
Add FDATransform (#1734) by @vectorvp

Switch to version-independent torchvision transforms.

If torchvision transforms v2 are available, they are used. Otherwise torchvision transforms v1 are used. For details see this comment.
Add Transform for DetCon + MultiViewTransformV2 for torchvision.transforms.v2 (#1737)

Typing, naming & docstring improvements

Type data/_utils(#1740), data/_helpers (#1742) and tests/models (#1744) by @vectorvp
Cleanup: docstrings in the lightly/data subpackage (#1741) by @ChiragAgg5k
Refactor: Update naming and remove unused package from AmplitudeRescaleTransform (#1732) by @vectorvp

Other

Fix DINOProjectionHead BatchNorm Handling (#1729)
Add masked average pooling for pooling with segmentation masks (DetCon)(#1739)

Many thanks to all of our contributors!

Models

- Python
Published by MalteEbner about 1 year ago

lightly - More Transforms, Typing and Docs Improvements

Changes

New transforms

Added RFFT2D and IRFFT2D transforms @snehilchatterjee
Add RandomFrequencyMaskTransform @payo101
Add GaussianMixtureMaskTransform @snehilchatterjee
Add AmplitudeRescaleTransform @payo101
Better support for both torchvision.transforms v1 and v2 without warnings/errors.

Added and updated docstrings

Many improvements by @Prathamesh010, @ayush22iitbhu, @ChiragAgg5k @HarshitVashisht11

Docs improvements

Improvements of the README.md @bhargavshirin and @kushal34712 @eltociear @Mefisto04 @ayush22iitbhu
Improvements of other parts of the the docs and tutorials @jizhang02
Fix examples on Windows @snehilchatterjee
Improve CONTRIBUTING.md @Prathamesh010
Added a back to top button for easier navigation @hackit-coder

More and better typing

Testing typing for all python versions
Typing of serve.py @ishaanagw
Cleanup: _image.py and _utils.py file in data subpackage @ChiragAgg5k

Better formatting

Move classes and public functions to top of file @fadkeabhi and @SauravMaheshkar

Other

Print Aggregate Metrics at End of Benchmarks as Markdown Table @EricLiclair

Many thanks to all of our contributors!

Models

- Python
Published by MalteEbner over 1 year ago

lightly - Support python 3.12

Support python 3.12, thanks @MalteEbner
update cosine warmup scheduler, thanks @guarin

Models

- Python
Published by japrescott over 1 year ago

lightly - enhanced library compatibility

Use TiCoTransform Everywhere
Refactor DINOLoss to not use center module
Add CenterCrop to val transform

Dependencies

Make library compatible with torch 1.10, torchvision 0.11, and pytorch lightning 1.6 (by using uv), thanks @guarin

Docs

Add notebooks, thanks @SauravMaheshkar
Add Timm Backbone Tutorial, thanks @SauravMaheshkar
Further docs and tutorial improvements

Models

- Python
Published by japrescott over 1 year ago

lightly - NumPy 2 support and new IBOTPatchLoss, KoLeoLoss

Added IBOTPatchLoss, KoLeoLoss and block masking, thanks @guarin
Allow learnable positional embeddings and boolean masking in masked vision transformer
Refactor IJEPA to use timm, thanks @radiradev

Dependencies

Allow NumPy 2, thanks @adamjstewart
Removed lightning-bolts dependency

Docs

Add finetuning tutorial, thanks @SauravMaheshkar
Fix MoCo link in DenseCL docs and further docs and tutorial improvements

Models

- Python
Published by japrescott over 1 year ago

lightly - DenseCL, SSL model improvements

Adds the DenseCL: Dense Contrastive Learning for Self-Supervised Visual Pre-Training method. See the docs.
Add TiCoTransform, thanks @radiradev!
Improvements to the pre-commit hooks, thanks @SauravMaheshkar!
Fix memory bank issue when using gather_distributed=True and training on a single GPU
Fix student head update in DINO benchmark
Various improvements to MaskedVisionTransformer
Renaming of Lightly SSL to LightlySSL

Models

- Python
Published by MalteEbner over 1 year ago

lightly - pydantic2 compatibility

Lightly is now compatible with pydantic2
migrated to pyproject

Models

- Python
Published by japrescott over 1 year ago

lightly - Numpy version compatibility

Two changes w.r.t numpy version 2: - Make lightly itself support numpy version 2: https://github.com/lightly-ai/lightly/pull/1561 - Disallow numpy 2.0 in the requirements, as torchvision is not yet compatible with numpy 2: https://github.com/lightly-ai/lightly/pull/1562

For more context, see https://github.com/lightly-ai/lightly/issues/1558

Models

- Python
Published by MalteEbner over 1 year ago

lightly - v1.5.7

Tiny improvements * Increase download timeout for json files (#1556) * Migrate coverage and mypy configuration to pyproject.toml #1549). Many thanks to @SauravMaheshkar for this improvement!

Models

- Python
Published by MalteEbner over 1 year ago

lightly -

Changes * Allow lightly-serve to run securely via https by passing sslcert and sslkey

Models

- Python
Published by japrescott over 1 year ago

lightly - Add unpatchify model utils operation

Changes * Add unpatchify model utils operation to reconstruct an image from its patches. See the PR for more information. Thanks to @randombenj for implementing this! * Fixes in CI regarding coverage. * Fixes in lightly-serve that the server was sometimes not shut down correctly.

Models

- Python
Published by MalteEbner over 1 year ago

lightly - Multi-GPU

Changes * Fixes the GatherLayer for multiple GPUs. See PR for more information. * Different typos in tutorials

Models

- Python
Published by japrescott almost 2 years ago

lightly - UX improvements

Changes * Removed the hydra warning when using lightly-serve * Improved the error messages and formatting of "well known" errors to improve the readability

Models

- Python
Published by japrescott almost 2 years ago

lightly - Better docs for MAE and TIMM

Changes * add benchmark results for MAE * add timm version info to docs

Models

- Python
Published by MalteEbner almost 2 years ago

lightly - MAE to use TIMM VIT

Changes * Refactor MAE to use TIMM VIT * Add AIM examples and docs * Updated BYOL and MOCO benchmarks * Use MMCRProjectionHead in examples

Models

- Python
Published by japrescott almost 2 years ago

lightly - Schedule Lightly Worker jobs with config version v4

This release includes some breaking changes for users of Lightly Worker.

Breaking Changes * Jobs are now scheduled with config v4 and require Lightly Worker 2.11 (breaking).

Changes * Add mmcr projection head (thanks @LukeSutor ) * Update argument type hints where the default is set to None to use Optional (thanks @otavioon) * Fix TiCoLoss (thanks @guarin ) * Add timm version check * fix parsing and caching issues with lightly-serve * allow to use lightly behind a proxy by setting HTTPS_PROXY and LIGHTLY_CA_CERTS

Models

- Python
Published by japrescott about 2 years ago

lightly - AIM Model and EMP-SSL Loss

Changes

Add EMP-SSL Loss (thanks @johnsutor).
Add AIM Model from Scalable Pre-training of Large Autoregressive Image Models.
- Benchmark code is here.
- Documentation is coming soon!
Add TiCo model code for ImageNet benchmark.
Add examples and documentation for MMCR loss (thanks @johnsutor).

Models

- Python
Published by ersi-lightly about 2 years ago

lightly - W-MSE Loss and Transform

Changes

Add MoCoV2 ImageNet benchmarks.
Make KNN feature normalization optional.
Implement W-MSE Loss and Transform (thanks @johnsutor).
Update generated specs with datasource expiration.

Models

- Python
Published by philippmwirth about 2 years ago

lightly - MMCR

Changes

Add MMCR loss and transform in https://github.com/lightly-ai/lightly/pull/1446. Thanks to @johnsutor!
Update README to reflect correct import for LightlyDataset in https://github.com/lightly-ai/lightly/pull/1437. Thanks to @dnth!
Fix MoCoV2 transform parameters in https://github.com/lightly-ai/lightly/pull/1441
Set model to eval for benchmark knn and linear classification in https://github.com/lightly-ai/lightly/pull/1444
Remove extra mean calculation in VICRegLoss in https://github.com/lightly-ai/lightly/pull/1450. Thanks to @RylanSchaeffer for pointing out the issue!
Fix MultiCropTransform cropmaxscales check in https://github.com/lightly-ai/lightly/pull/1454. Thanks to @Djoels for pointing out the issue!

Models

- Python
Published by guarin about 2 years ago

lightly - Typecheck part of models

Changes

Typecheck part of models
Polish benchmarks page
Update specs

Models

- Python
Published by michal-lightly about 2 years ago

lightly - Typecheck part of utils

Changes

Typecheck part of utils
Add ability to download the new reportv2.json
Update specs

Models

- Python
Published by japrescott over 2 years ago

lightly - Cyclic Cosine Scheduler

Changes

Add period argument to CosineWarmupScheduler #1413, thanks to @JannikWirtz!

Models

- Python
Published by guarin over 2 years ago

lightly - Documentation Improvements

Changes

Small improvements to documentation
Added types to embedding scripts (kudos to @agpeshal!)
Fix: Handle OPTIONS method for lightly-serve script

Models

- Python
Published by michal-lightly over 2 years ago

lightly - VICReg benchmarks on Imagenet

Changes

We have evaluated our VICReg implementation on Imagenet (check it out).
Docs update to emphasize difference between Lightly SSL and the company.
Allow filenames with commas in embedding files and datasets.
Fix: Imagenet benchmarks memory problems.

Models

- Python
Published by philippmwirth over 2 years ago

lightly - Types!

Changes

Add mypy and type the package partially (#1382). lightly.transforms is fully typed. We'll gradually add types for the other modules.
Add py.typed files for typed parts of the package (#1382). This makes types available when working with lightly from other codebases.
Add support to resume benchmark training (#1347). Thanks a lot to @sadimanna!
Remove docs for outdated/internal API methods (#1385).
Make the relative_path argument optional when scheduling a Lightly Worker run with local storage (#1384).

Models

- Python
Published by guarin over 2 years ago

lightly - BarlowTwins Benchmark on ImageNet

Changes

Added a new benchmark of BarlowTwins on ImageNet.
Optimized performance of the BarlowTwinsLoss computation, making it much faster
Fixed a bug in the CosineWarmupScheduler. Thanks to @anishacharya for pointing out the problem.
Cleaned the setup.py

Models

- Python
Published by MalteEbner over 2 years ago

lightly - Lightly Local Workflow

Changes

Prepare for local workflow support:
- add lightly-serve command
- regenerate specs
Fix docstrings

Models

- Python
Published by japrescott over 2 years ago

lightly - Patch generated API client

Changes

Patch generated API client.

Models

- Python
Published by philippmwirth over 2 years ago

lightly - BYOLTransform

Changes

Add BYOLTransform which replaces SimCLRTransform in BYOL benchmarks.
Log benchmark results only on rank0.
Fix bug in PMSNLoss where probabilities were not converted to log-space before the loss calculation. Thanks to @Cloudy1225 for reporting this!
Version check now runs in background and no longer requires SIGALRM.
Add support for scheduling Lightly Worker runs with the new selection strategy strength option.

Models

- Python
Published by guarin over 2 years ago

lightly - I-JEPA

Changes

Basic support for I-JEPA (thanks to @Natyren!)
add BYOL imagenet resnet50 benchmark

Models

- Python
Published by japrescott over 2 years ago

lightly - Paginate API client endpoints

Changes

Paginate API client endpoints
Cleaned up API client codebase

Models

- Python
Published by michal-lightly over 2 years ago

lightly - Remove prefetch-generator

Changes

Remove prefetch_generator as this is supported natively by PyTorch 1.7 and higher.
Improve error messages for scheduled jobs with invalid configurations.
Correctly create SelectionConfig with repeated object references.

Models

- Python
Published by philippmwirth over 2 years ago

lightly - SimCLR with finetuning benchmark

Changes

updates SimCLR benchmark results with finetuning added
fixes swav imagenet benchmark
it is now possible to fetch team datasets

Models

- Python
Published by shaundaley39 over 2 years ago

lightly - DINO ResNet50 benchmarks

Changes

Add benchmarks of the DINO model using a ResNet50 backbone #1254
openapi: setuptools not required at run-time #1289 Thanks to @adamjstewart for making this fix!

Models

- Python
Published by MalteEbner over 2 years ago

lightly - New API Generator

Changes

Update generated API code to use openapi-generator instead of swagger-codegen (#1271, #1275, #1276, #1281)
- This allows use to better validate API requests and detect issues earlier
Update tutorials to use transforms instead of collate functions #1277
Add reference in README to the Reverse Engineering Self-Supervised Learning paper which uses Lightly #1278
Fix examples to work with LightlyDataset #1280

Models

- Python
Published by guarin over 2 years ago

lightly - Benchmarks and deprecated code removal

Changes

SwAV ImageNet benchmark have been added (#1231)
VicRegL Vertical Flip Probability (#1266) thanks to @iamharsha1999
The multiviewcollate is now replaced with default collate (#1262)
Docstrings have been updated
We have removed deprecated code for active learning (#1251) and recommend to use our more powerful approach to active learning as outlined in one or our tutorials
We have removed deprecated lightly-upload code (#1247)

Models

- Python
Published by japrescott over 2 years ago

lightly - Small fixes

Changes

Fix torch.distributed import in vicregl_loss.py (#1239)
API Client: remove default querytagname from ApiWorkflowClient.upload_scores (#1243)

Models

- Python
Published by huan-lightly-0 almost 3 years ago

lightly - SimCLR ImageNet benchmark and improved documentation

Changes

add benchmark for SimCLR ImageNet
API client improvements: sample creation with write URLs, handling of relevant filenames, support for num_processes and num_threads
documentation improvements: a cleanup of the README, new overview chart and better links

Models

- Python
Published by shaundaley39 almost 3 years ago

lightly - MocoV3 and better support for distributed

Changes

add support for MoCoV3 projection head. Thanks to @adamjstewart
better support for distributed training: Better error messages and bugfixes
updated benchmark results on Cifar10

Models

- Python
Published by MalteEbner almost 3 years ago

lightly - SimCLRV2 Projection Head

Changes

Add support for SimCLRV2 projection head. Thanks to @adamjstewart!
Add by default BatchNorm layers to SimCLR projection head. Thanks to @adamjstewart!

Models

- Python
Published by guarin almost 3 years ago

lightly - pytorch 2.0 compatibility

Changes

Lightly is now compatible with PyTorch 2.0 (but Lightly itself does not use it, yet)
It's now possible to install Lightly "lightly" by only installing the parts necessary for API communication
Support newer setuptools (Thanks @adamjstewart ! )
Added missing config options
Improved docstrings and document potential API errors

Models

- Python
Published by japrescott almost 3 years ago

lightly - FastSiam

Changes

New FastSiam model: FastSiam: Resource-Efficient Self-supervised Learning on a Single GPU
Add helper to list registered Lightly Workers

Models

- Python
Published by philippmwirth almost 3 years ago

lightly - Schedule Lightly Worker jobs with config version v3

This release includes some breaking changes for users of Lightly Worker.

Breaking Changes - Jobs are now scheduled with config v3 for Lightly Worker 2.6 (breaking). - Remove object_level config option (breaking).

Changes - Automate release using Github actions - Split ApiWorkflowClient download and export functionality - Preparation for instance segmentation support

Models

- Python
Published by michal-lightly almost 3 years ago

lightly - Prior Matching for Siamese Networks

Changes

New PMSN model: Prior Matching for Siamese Networks, 2022.
Add deprecation warning for active learning workflow.
Add deprecation warning for collate functions.
Remove deprecated documentation.
Refactor use of transforms.

Models

- Python
Published by shaundaley39 almost 3 years ago

lightly - Pytorch Lightning 2.0 Compatibility and fixes

Pytorch Lighting 2.0 Compatibility

Pytorch Lightning introduced breaking changes in the ways devices and accelerators are specified. We updated the code and example models to reflect that. For details, see the PR.

Benchmarks now use transforms

The benchmarks now do the augmentations (e.g. colour jitter) in the dataset transform instead of the collate function of the dataloader and have been updated. For details, see the PR

Other changes.

The ApiWorkflowClient is now pickable, improving multithreading capabilities.
The LabelBox export now supports LabelBox format v4.
Smaller fixes for a better user experience.

Models

- Python
Published by MalteEbner almost 3 years ago

lightly - Better Lightly Worker job validation and fixes

Changes

Raise error when trying to schedule a Lightly Worker job with unknown configuration arguments.
Add get_compute_worker_run_checkpoint_url method to ApiWorkflowClient, allowing to access a pretrained checkpoint from the Lightly Worker by URL.
Fix error in Lightly version check on Windows.
Remove deprecated PytorchLightning progress_bar_refresh_rate trainer argument in tutorials.
Make Masked Autoencoder work with half-precision training.

Other

The lightly package is now formatted using the black code formatter and isort.

Models

- Python
Published by guarin almost 3 years ago

lightly - torch(vision) and lightly.api independence

This release includes some breaking changes, especially for our users of the the Lightly Worker Please follow the migration guide to see version compatibility.

Breaking Changes

Make lightly.api independent of torch(vision) (breaking).
Validate the config created by the api (breaking).
Fix scheduling jobs with config v2 (breaking).

Changes

Add benchmarks results with new GaussianBlur implementation.
Add transforms for all SSL models.
Remove extra pooling layers from benchmarks.

Other

Set creator on various endpoints in pip.

Models

- Python
Published by japrescott almost 3 years ago

lightly - Deprecate Gaussian blur scale and kernel_size

Changes

Add deprecation warning for GaussianBlur scale and kernel_size.
Raise an error for an unsuccessful json download.
Enable OBS datasource configurations.
Update generated api code.

Models

- Python
Published by philippmwirth about 3 years ago

lightly - Add SwaV Queue benchmarks on Imagenette

Changes

Add SwaV Queue benchmarks on Imagenette
Ability to pass dataset creator to Lightly API

Models

- Python
Published by michal-lightly about 3 years ago

lightly - SwaV architecture docs

Changes

Improved SwaV architecture documentation and an example
Unit testing of frozen prototypes
Gaussian blur implementation fix

Models

- Python
Published by shaundaley39 about 3 years ago

lightly - Better SSL schedulers, benchmarks for TiCo and VicRegl

Changes

We added cosine decay for the DINO, BYOL, and MOCO models.
We added a CosineWarmupScheduler class in the benchmarks.
We added benchmark results for the TiCo and VICRegL models.

Models

- Python
Published by MalteEbner about 3 years ago

lightly - Deprecate `lightly-upload`

Changes

Deprecate dataset upload via CLI (lightly-upload). In the future, only uploads with the Lightly Worker will be supported.
Add SimMIM and TiCo benchmarks.

Models

- Python
Published by guarin about 3 years ago

lightly - TiCo and VICRegL

Changes

New TiCo model.
New VICRegL model.
API Client extended with listdatasourcepermissions.
Bug fix generalizing SMoGProjectionHead. Thanks @DrIsDr!
Added version compatibility check.

Models

- Python
Published by shaundaley39 about 3 years ago

lightly - Prototype Freezing for SwaV

Changes

Add SwAV prototype freezing. Thanks @ibro45 for the contribution!
Add function to download embeddings.
Update links in the README to point to the new docs.

Models

- Python
Published by philippmwirth about 3 years ago

lightly - Queue for SwAV

Changes

Add queue implementation for SwAV. Thanks @ibro45 for the contribution!
Fix memory bank initialization
Fix API client version compatibility check

Models

- Python
Published by michal-lightly about 3 years ago

lightly - SimMIM

SimMIM

We added the SimMIM model. It has a very similar architecture to MAE, but it uses a ViT encoder using both masked and non-masked patches as input. Furthermore it has just a simple linear layer as a decoder and uses L1 instead of L2 loss.

Models

- Python
Published by MalteEbner about 3 years ago

lightly - Update API Client & Fix Citations

Changes

Fix citations in README.md (Special thanks to @utkuozbulak)
Update generated API code for Lightly API

Models

- Python
Published by guarin about 3 years ago

lightly - VICReg

Changes

Added VICReg Model (Special thanks to Boris Albar @b-albar). (See VICReg example)
The memory bank now works with distributed training.
Speedup all code working with runs and datasets.
Add helpers for predictions.

Models

- Python
Published by japrescott about 3 years ago

lightly - Imagenet100 benchmarks and more random rotations

Changes

Added Imagenet100 benchmarks
Allow more finegrained control over random rotations

Models

- Python
Published by philippmwirth about 3 years ago

lightly - Update API Client

Changes

Update API client to use RunWorkerLabels
Updated API client documentation for tag endpoints

Models

- Python
Published by MalteEbner over 3 years ago

lightly - Update API Client

Changes

Update API client embedding upload

Models

- Python
Published by shaundaley39 over 3 years ago

lightly - Update API Client & Lightly Worker Docs

Changes

Update API client for Lightly Worker v2.3.7
Add docs for artifacts uploaded from the Lightly Worker
Add docs for advanced relevant filenames options for the Lightly Worker
Update docs for diversity selection strategy input types

Models

- Python
Published by guarin over 3 years ago

lightly - Update Lightly Worker Docs & API Client

SMog Example

A simple example demonstrating the usage of SMog was added. Otherwise no major changes have been made

Other Changes

Documentation updates regarding datapool, using predictions and subdir information
Allow self signed certificates if API is self-hosted
Utilities to work with artifacts

Models

- Python
Published by japrescott over 3 years ago

lightly - Multi-prototype SWaV

Multi-prototype SWaV (thanks a lot @Atharva-Phatak 🙂)

Implements a multi-prototype head for SWaV as discussed in https://github.com/lightly-ai/lightly/issues/944.

Other Changes

Fixes typehints for export functions in the ApiWorkflowClient
Documentation updates
Add scores argument to ObjectDetectionOutput

Models

- Python
Published by philippmwirth over 3 years ago

lightly - SMoG

SMoG: Unsupervised Visual Representation Learning by Synchronous Momentum Grouping

Lightly 1.2.30 comes with the SMoG model introduced in Unsupervised Visual Representation Learning by Synchronous Momentum Grouping. Documentation and benchmarks will be released soon.

Breaking Change

in the ApiWorkflowClient, create_dataset now throws an error where a dataset of the same name already exists. To reuse an existing dataset users should switch to using setdatasetidbyname.

Other Changes

OBS (object storage service) remote datasources now supported
documentation improvements

Models

- Python
Published by shaundaley39 over 3 years ago

lightly - Improved Active Learning Score computation

Improved Active Learning Score computation

We slightly refactored the Active Learning Score computation to make it better from a software development point of view.

Models

- Python
Published by MalteEbner over 3 years ago

lightly - Masked Siamese Networks

MSN: Masked Siamese Networks for Label-Efficient Learning

Lightly 1.2.28 comes with the new MSN model introduced in Masked Autoencoders Are Scalable Vision Learners. Please head over to our docs to see how to use MSN with Lightly: https://docs.lightly.ai/examples/msn.html

Other Changes

Lightly is now compatible with Pytorch Lightning v1.7

Models

- Python
Published by guarin over 3 years ago

lightly - Server Side Encryption + Documentation

Changes

if server side encryption is enabled for s3 datasources, the proper headers are sent
expose advanced selection configuration classes and enums for typed configurations

Documentation

Clarify which permissions need to be set for GCS when running a datapool
Add instructions on how to re-use a checkpoint
Minor docs updates

Models

- Python
Published by japrescott over 3 years ago

lightly - Patch download speed and dependency issue

Changes

Speed up lightly-download using multithreading
Documentation updates
Hotfix: Fix compatability issues with pytorch-lightning>=1.7

Models

- Python
Published by philippmwirth over 3 years ago

lightly - API update for working with delegated access

Api connections

Delegated access use lightly urls by (#875)

Models

- Python
Published by IgorSusmelj over 3 years ago

lightly - Delegated Access and Documentation Updates

Documentation

new account ID for delegated access (#872)
docs around loading model from Lightly worker model checkpoint (#870)
additional speedup information around maxepochs and numworkers settings (#873)
improved README (#871)
datasource documentation udpated (#867)

Testing

tox tests fixed (#869)
removed test assertion invalid for lower torchvision versions (#865)

Dependencies

pyav version relaxed (#868)

Models

- Python
Published by shaundaley39 over 3 years ago

lightly - Documentation updates and API connections

Documentation

Improved docs for active learning (#862)

Api connections

Datasource loading now allows to use a tqdm progress bar (#860)
All API requests now have a timeout (#863)
Video downloads also have a timeout (#864)

Models

- Python
Published by MalteEbner over 3 years ago

lightly - Documentation Updates and Improved Video Loading

Documentation

New docs on how to create frame predictions compatible with the Lightly platform (#857)
New docs for sequence selection features in the Lightly worker (#856)
Remove duplicated section in docs (#855)
Updated docs for first steps with the Lightly worker (#858)

Video Loading

Fixed loading of videos with wrong metadata (#853)

Other

Removed trailing comma in filenames exported from API (#859)

Models

- Python
Published by guarin over 3 years ago

lightly - PIRL and more API helpers

PIRL and more API helpers

Self-Supervised Learning of Pretext-Invariant Representations

Support for the PIRL collate function has been added (#850). Special thanks to @shikharmn for contributing this!

Improvement

Expose functionality to export the filenames of the samples within a tag (#852)
Better error handling of requests by passing sessions (#851)

Models

- Python
Published by japrescott over 3 years ago

lightly - Refresh docs and more API helpers

Refresh docs and more API helpers

Docs

@jwuphysics noticed and fixed some typos in the docs, thanks a lot!
@MalteEbner found some more and fixed them too 🙂

Support for role based access to S3 from ApiWorkflowClient

With #841 we added helpers to configure a Lightly dataset with delegated access rules.
#847 added the necessary documentation

Models

- Python
Published by philippmwirth over 3 years ago

lightly - Examples for MAE and better docs

Masked Autoencoders Examples

We added examples for the MAE model

Docs

We added docs for the collapse detection helper
We added docs for plotting positive and negative example images

Models

- Python
Published by IgorSusmelj over 3 years ago

lightly - Masked Autoencoder, SSL debug utilities

Masked Autoencoders

We implemented the paper Masked Autoencoders Are Scalable Vision Learners. https://arxiv.org/abs/2111.06377 is suggesting that a masked auto-encoder (similar to pre-training on NLP) works very well as a pretext task for self-supervised learning. See https://github.com/lightly-ai/lightly/issues/721 for more details. Thanks to @Atharva-Phatak for helping us figure out a good implementation method.

Collapse detection helper for SimSiam

We added a helper for detecting a collapsing SimSiam network. See https://ar5iv.labs.arxiv.org/html/2011.10566#S4.SS1 for more details.

Plot positive and negative example images

We added a helper to plot positive and negative example images, which also allows seeing what the augmentations do. See https://github.com/lightly-ai/lightly/pull/818 for more details.

Models

- Python
Published by MalteEbner over 3 years ago

lightly - Video Dataset initialisation improvements

Video Dataset initialisation speedup

When initialising a LightlyDataset on a directory with videos, all frames in all videos have to be counted to know the number of frames in the dataset and their filenames. This process now uses multihreading over videos and can thus be much faster.

Video Dataset initialisation bugfix

We fixed a bug that the number of frames was estimated wrongly based on the length of the video when using the pyav backend.

Video Dataset initialisation progress bar

When initialising the video dataset, a progress bar over the videos is shown. This is helpful information for datasets with many videos.

Models

- Python
Published by MalteEbner almost 4 years ago

lightly - Decoupled Contrastive Learning, Model heads with default parameters

Decoupled Contrastive Learning

We have implemented the Decoupled Contrastive Learning (DCL) loss. It allows faster training with smaller batch sizes than other self-supervised learning models.

Documentation: https://docs.lightly.ai/examples/dcl.html Decoupled Contrastive Learning paper: https://arxiv.org/abs/2110.06848

Model heads with default parameters

All model heads have now default parameters following the values of the original papers.

Create custom metadata config

Custom metadata in the Lightly Platform can now easily be configured via ApiWorkflowClient.create_custom_metadata_config()

Progress bar for video dataset initialization

Constructing a LightlyDataset with large video datasets can take long, as all frames in all videos have to be counted. We added a tqdm progress bar for it.

Models

- Python
Published by MalteEbner almost 4 years ago

lightly - More stable image and video downloads

More stable image download

Lightly now also retries an image download if the read url exists at first, but becomes unavailable during the download.

Models

- Python
Published by MalteEbner almost 4 years ago

lightly - Improved Video Loading

Improved Video Loading

Loading frames from videos was updated and we now verify that the timestamp of the loaded frame is correct. This fixes a bug where sometimes a frame with a different timestamp than the one requested by the user was loaded. Lightly will now also warn you if the correct frame could not be found.

Documentation Updates

We added many documentation updates on how you can use and interact with the Lightly Platform.

Models

- Python
Published by guarin almost 4 years ago

lightly - SimSiam and bug fixes

SimSiam and bug fixes

SimSiam fixes

@gergopool noticed and fixed two minor issues with the SimSiam implementation! Thanks a lot 🙂

Bug fix: Embedding doesn't work with prefetch generator

Fixed the bug introduced with #748 that the embedding images with lightly didn't work when the prefetch_generator package is installed.

Models

- Python
Published by philippmwirth almost 4 years ago

lightly - Improved embedding progress bar, Label export, Docs improvements

Improved embedding progress bar, Export to Label, Docs improvements

Improved embedding progress bar

The embedding progress bar now shows the number of embedded images instead of embedded batches.

Label export

You can now use the ApiWorkflowClient to export data from the LightlyPlatform as a .json file to be imported into a label tool.

Docs improvements

Some errors in the docs have been fixed.

Models

- Python
Published by MalteEbner almost 4 years ago

lightly - DINO tricks, Imagenette benchmarks, API improvements

DINO tricks, Imagenette benchmarks, API improvements

DINO tricks

DINOHead now allows to freeze the last layer which stabilizes the model performance. DINOHead now also allows to normalise the last layer. This was implemented by @Atharva-Phatak. Thank you very much!

Imagenette benchmarks

We now include benchmarks of all models on Imagenette.

API Improvements

Better documentation of custom metadata

The CLI command to upload custom metadata is now included in the command line tool examples.

Better dataset upsizing

Upsizing a dataset in the Lightly Platform by adding more samples to it now cannot happen accidentally anymore, instead you have to specify append=True. Furthermore, bugs regarding appending new custom metadata have been fixed.

Create ApiWorkflowClient with token from env

When creating an ApiWorkflowClient, you can now pass the token as environment variable LIGHTLY_TOKEN instead of as argument.

Bugfixes in `check_embeddings()`

When checking embedding files, now columns like masked and selected are accounted for properly.

Models

- Python
Published by MalteEbner almost 4 years ago

lightly - Active Learning for Keypoint Detection and New API Workflows

Active Learning for Keypoint Detection and New API Workflows

Active Learning: Keypoint Detection

#707 added an active learning scorer for keypoint detection. This means that you can now do active learning for keypoint detection with Lightly!

New API Workflows

With #712 and #714 we added all the necessities to configure and run the compute worker from your Python code.

Access Shared Datasets

722 makes it possible to access datasets which were shared with you in the Lightly web-app from your Python code as if they were your own.

Documentation

We made many smaller changes and fixes in the docs 🙂

Models

- Python
Published by philippmwirth almost 4 years ago

lightly - Bugfixes and Documentation Improvements

Bugfixes and Documentation Improvements

Bugfix in trainmodelandembedimages()

Fixed the bug that the train_model_and_embed_images() function did not return the embeddings, filenames and labels. Thanks for finding this bug @opassos!

Bugfix in video loader

When using the video_loader backend of torchvision, it could fail when training for multiple epochs and had a linearly growing demand for memory and file handlers. This was fixed.

Documentation update in Readme.md

The benchmark table for CIFAR10 was fixed and the DINO model added to the list of supported models.

Models

- Python
Published by MalteEbner almost 4 years ago

lightly - DINO and Updated Benchmarks

DINO:

We added support for the DINO model which works with ResNet and Transformer backbones! To get started head over to our examples section in the docs which contains example code for DINO using PyTorch and PyTorch Lightning.

Update Cifar10 Benchmarks

We updated the Cifar10 benchmarks with results for all models provided by Lightly. The benchmark does now only use low level blocks which we introduced in a previous release. You can find the updated benchmark results and code in our docs.

Models

- Python
Published by guarin almost 4 years ago

lightly - Faster Upload and more API methods

Faster Image Upload

When uploading images through the CLI interface we now by default use 8-32 threads to speed up the process. This setting can be controlled with using the loader.num_workers argument.

Create Tags from Filenames

The ApiWorkflowClient now has a create_tag_from_filenames method which allows you easily to create a new tag on the Lightly Platform by passing a list of filenames.

Updated Visibility of Environment Variables in Docs

All environment variables in the docs should now be properly indicated using braces: {ENV_VARIABLE}

Models

- Python
Published by guarin about 4 years ago

lightly - # Bug Fixes for Video datasets

Bug fixes for Video datasets

Make LightlyDataset threadsafe for videos

The VideoLoader now uses threading.local which allows the loader to be used from multiple threads.

LightlySubset of VideoDataset returns correct filenames

Resolved the bug when calling subset.get_filenames() on the subset of a video dataset would return the filenames of the full dataset instead of the subset.

Download of videos allows restarting

We added retries to handle errors when downloading videos.

Getting video meta information allows ignoring metadata

When trying to find out meta information about a video like the number of frames or timestamps there are two ways: - Try to read it from the metadata. However, sometimes this metadata is missing or false. - Find it out by reading the video itself. This is computationally expensive

We now allow ignoring metadata and choosing the 2nd option, which is slower but guaranteed to give the correct data.

Models

- Python
Published by MalteEbner about 4 years ago

lightly - Bug fixes

Bug fixes

Active learning scores from generators

Raise an error if the user intends to use active learning scores but they don't exist.

Clone labels to free file handlers

Resolved the bug where users running lightly-magic on large datasets have gotten errors because of too many open files.

Wrap requests to S3 in retries

We added retries to handle 5XX errors according to best practices.

Models

- Python
Published by philippmwirth about 4 years ago

lightly - Multi-GPU support, format check for embedding file

Multi-GPU support, format check for embedding file

Multi-GPU

We now support using lightly with multiple GPUs. For reference, look at the docs.

Format check for embedding file

When trying to upload an embedding file to the Lightly Platform it is now checked to have the correct format: It must have the column names in the correct order and without whitespaces and must not have empty rows.

Models

- Python
Published by MalteEbner about 4 years ago

lightly - Faster metadata upload, bugfixes and improvements

Faster metadata upload, bugfixes and improvements

Faster metadata upload

The upload of custom metadata is now done async and with multiple workers in parallel, allowing to speed up the upload process by up to 30 times.

Bugfixes

When there is a failure uploading a file to a signed url, now the status code is printed correctly.
Creating a LightlyDataset with an input_dir with videos will now raise all errors scanning the input directory instead of ignoring them. This means e.g. that if a subfolder without read permissions is encountered, a PermissionError will be raised instead of silently ignoring the subfolder.
When embedding, the order of the embeddings in the output will now be the order of the samples in the dataset, even if multiple workers are used in the dataloader. Thus also the embeddings in the embedding file are in the sorted order. This is not directly a bugfix but might prevent problems later on.

Improvements

The usage of resnet backbones in the example models is now consistent. Thanks for bringing this up @JeanKaddour!
The SimCLR example now does not use Gaussian blur anymore, just like in the paper. Thanks for pointing this out @littleolex!
The BarlowTwins example now also uses an input size of 32 to make it consistent with the other examples. Thanks for bringing this up @heytitle!
The documentation for setting up Azure as cloud storage for the Lightly Platform has been improved. ## Models
Bootstrap your own latent: A new approach to self-supervised Learning, 2020
Barlow Twins: Self-Supervised Learning via Redundancy Reduction, 2021
SimSiam: Exploring Simple Siamese Representation Learning, 2020
MoCo: Momentum Contrast for Unsupervised Visual Representation Learning, 2019
SimCLR: A Simple Framework for Contrastive Learning of Visual Representations, 2020
NNCLR: Nearest-Neighbor Contrastive Learning of Visual Representations, 2021
SwAV: Unsupervised Learning of Visual Features by Contrasting Cluster Assignments, M. Caron, 2020

- Python
Published by MalteEbner about 4 years ago

lightly - Speeding up io and improving documentation

Bug fixes and documentation updates

Documentation Updates

We have now added support for other cloud storage providers. You can now work directly with data stored in AWS S3, Azure Blob Storage, and Google Cloud Storage. Furthermore, you can stream data directly from your local filesystem in the Lightly Platform without uploading any images/ videos to any cloud. Check out the instructions here!

Performance

We improved the dataset indexing that is used whenever you create a lightly dataset. Indexing of large datasets (>1 mio samples) now works much faster.

Models

- Python
Published by IgorSusmelj about 4 years ago

Recent Releases of lightly

lightly - v1.15.22

What's Changed

New Contributors

Models

lightly - v1.15.21

What's Changed

New Contributors

Models

lightly - v1.5.20

Changes

Models

lightly - v1.5.19

Updates to the Model Documentations and addition of MACL Loss

Changes

Models

lightly - update SimCLR and MAE model docs

What's Changed

Models

lightly - cleanup & detcon loss fix

Changes

Models

lightly - DetCon and typings

Changes

Typing and Docs

Models

lightly - More and better Transforms

Changes

New transforms

Switch to version-independent torchvision transforms.

Typing, naming & docstring improvements

Other

Models

lightly - More Transforms, Typing and Docs Improvements

Changes

New transforms

Added and updated docstrings

Docs improvements

More and better typing

Better formatting

Other

Models

lightly - Support python 3.12

Models

lightly - enhanced library compatibility

Dependencies

Docs

Models

lightly - NumPy 2 support and new IBOTPatchLoss, KoLeoLoss

Dependencies

Docs

Models

lightly - DenseCL, SSL model improvements

Models

lightly - pydantic2 compatibility

Models

lightly - Numpy version compatibility

Models

lightly - v1.5.7

Models

lightly -

Models

lightly - Add unpatchify model utils operation

Models

lightly - Multi-GPU

Models

lightly - UX improvements

Models

lightly - Better docs for MAE and TIMM

Models

lightly - MAE to use TIMM VIT

Models

lightly - Schedule Lightly Worker jobs with config version v4

Models

lightly - AIM Model and EMP-SSL Loss

Changes

Models

lightly - W-MSE Loss and Transform

Changes

Models