v0.13.2
Added
- Support for pretraining RF-DETR 1.3 models.
Changed
- Export only EMA weights for object detection models. This reduces the exported model
size by 2x.
Fixed
- Fix pretrained ViT-small panoptic segmentation model checkpoint.
What's Changed
- Support new RF-DETR 1.3 models (Small, Nano, Medium, SegPreview) by @masakljun in #503
- Update DINOv3 ViT-S Panoptic COCO Checkpoint by @guarin in #519
- Update events for predictions by @IgorSusmelj in #521
- Export only EMA model by @guarin in #516
- Add DINOv3 fine-tuning tests by @guarin in #517
- Release 0.13.2 by @guarin in #524
Full Changelog: v0.13.1...v0.13.2
v0.13.1
Fixed
- Fix bug in ONNX export for object detection models.
What's Changed
- Add argument for number of channels and better inference. by @stegmuel in #512
- Release v0.13.1 by @guarin in #515
Full Changelog: v0.13.0...v0.13.1
v0.13.0
New DINOv3 Tiny Object Detection Models: We release tiny DINOv3 models pretrained on
COCO for object detection!
New DINOv3 Panoptic Segmentation: You can now run inference and fine-tune DINOv3 models
for panoptic segmentation!
Added
- New COCO pretrained tiny LTDETR models
vitt16andvitt16plus. - Support for DINOv3 panoptic segmentation inference and fine-tuning.
- Quick start guide for object detection.
- Possibility to load backbone weights in LTDETR.
- ONNX export for LTDETR.
- Add Weights & Biases logging support for all fine-tuning tasks.
- Log best validation metrics at the end of training.
Changed
- Rename
lightly_train.train()tolightly_train.pretrain(). The old name is still
available as an alias for backward compatibility but will be removed in a future release. - Restructured the documentation to better reflect the different workflows supported
by LightlyTrain.
Fixed
- Fix bug in
model.predict()for object detection models. - Fix bug in object detection transforms when using images with dtype float32.
- Fix bug when running pretraining on an MPS device.
- Fix bug when resuming training with a recent PyTorch version.
- Fix bug when resuming a crashed run that was initialized from a pretrained COCO model.
What's Changed
- add log of best result by @masakljun in #475
- Thomas trn 1628 small od models by @stegmuel in #430
- Object detection support backbone weights by @stegmuel in #483
- Fix resuming from coco init by @stegmuel in #485
- Add panoptic segmentation by @guarin in #488
- Load teacher on correct device by @guarin in #490
- Thomas trn 1653 ltdetr onnx export by @stegmuel in #487
- Remove RSNA/Neurips banner by @stegmuel in #494
- Fix resuming error with recent torch versions by @guarin in #492
- Fix learning rate monitor float64 MPS issue by @guarin in #496
- Remove teacher eval mode warning by @guarin in #495
- Masa trn 1695 load ema weights only by @masakljun in #484
- Use latency everywhere by @stegmuel in #498
- Add links to vit tiny checkpoints by @stegmuel in #497
- Fix pytest out of disk space by @guarin in #502
- Cleanup transforms dtypes and asserts by @guarin in #499
- Revert docs CSS changes by @guarin in #504
- Restructure docs by @guarin in #489
- Update distillation quickstart by @guarin in #506
- Release v0.13.0 by @guarin in #507
New Contributors
- @masakljun made their first contribution in #475
Full Changelog: v0.12.4...v0.13.0
v0.12.4
Fixed
- Fix bug in
model.predict()for object detection models.
What's Changed
Full Changelog: v0.12.3...v0.12.4
v0.12.3
Added
- Add support for specifying data configs in YAML format.
Changed
- Improve the layout of the object detection training logs.
Deprecated
- Deprecate
reuse_class_headargument intraincommand. The model will now
automatically reuse the classification head only when the number of classes in the
data config matches that in the checkpoint. Otherwise, the classification head will
be re-initialized.
Fixed
- Fix
image_sizenot tuple when training from pretrained model. - Fix a bug when fine-tuning a model with
resume_interrupted=True. - Fix
num_classesnot updated when loading an object detection checkpoint with
different number of classes.
What's Changed
- Fix and speed up unit tests by @guarin in #450
- Add events for training detection and segmentation models by @IgorSusmelj in #447
- Fix dist training issues by @yutong-xiang-97 in #446
- Fix resume interrupted for fine-tuning by @guarin in #448
- Support yaml data config by @yutong-xiang-97 in #455
- Improve Metrics Logging by @yutong-xiang-97 in #458
- Internal class id mapping in Object Detection dataset by @yutong-xiang-97 in #462
- Update docs/readme by @guarin in #454
- Fix url of SAT493m url by @stegmuel in #465
- Deprecate reuse class head and use hooks for segmentation by @yutong-xiang-97 in #466
- Add RSNA/NeurIPS README banner by @liopeer in #468
- Fix README title by @liopeer in #469
- Fix OD
num_classesissue. by @yutong-xiang-97 in #467 - Release v0.12.3 by @guarin in #472
Full Changelog: v0.12.2...v0.12.3
v0.12.2
Fixed
- Fix
image_sizenot tuple when training from pretrained model.
What's Changed
- Fix object detection image size from checkpoint by @guarin in #444
- Release v0.12.2 by @guarin in #445
Full Changelog: v0.12.1...v0.12.2
v0.12.1
Added
- Add support for DINOv3 instance segmentation
inference and fine-tuning. - Add support for loading DICOM images
as input data for training and inference. - Add event tracking, disable with
LIGHTLY_TRAIN_EVENTS_DISABLED=1 - Add support for fine-tuning object detection models with custom image resolutions.
What's Changed
- Add object detection colab by @guarin in #418
- Add SAT493m models to distillation docs by @liopeer in #421
- Add DICOM image loading support by @yutong-xiang-97 in #416
- Thomas trn 1628 small vit by @stegmuel in #417
- Save object detection prediction results in COCO format by @yutong-xiang-97 in #422
- Add event tracking by @IgorSusmelj in #406
- Instance segmentation improvements by @guarin in #423
- Add
init_weightsto DINOv3 ViT to fix the flaky tests by @yutong-xiang-97 in #427 - Add eomt instance segmentation docs by @guarin in #426
- Update detection image by @guarin in #429
- Freeze teacher parameters in Distillation and DistillationV2 by @OshriFatkiev in #434
- Document Non-RGB & DICOM loading support by @yutong-xiang-97 in #432
- Disable unit test tracker by @guarin in #438
- Add object detection custom image resolution fine-tuning by @guarin in #437
- Release v0.12.1 by @guarin in #440
New Contributors
- @IgorSusmelj made their first contribution in #406
- @OshriFatkiev made their first contribution in #434
Full Changelog: v0.12.0...v0.12.1
v0.12.0
💡 New DINOv3 Object Detection: Run inference or fine-tune DINOv3 models for object detection! 💡
Added
- Add support for DINOv3 object detection model training.
- Add semantic segmentation autolabeling support with
predict_semantic_segmentation. - Add support for DINOv3 ConvNeXt models.
- Automatically download DINOv3 weights.
- Add support for passing pretrained model names or checkpoint paths as
modelargument
to the model training functions liketrain_semantic_segmentation.
Changed
- Widen PyTorch constraint — remove
<2.6upper bound to allow PyTorch 2.6 and later that is officially supported by PyTorch Lightning 2.5. - Rename
load_model_from_checkpointtoload_model. The function now downloads
checkpoints that do not exist locally.
Fixed
- Fix issue with loading DINOv3 SAT493m checkpoints.
- Fixed an issue where dataset cache files were incorrectly saved.
What's Changed
- no model name in predictions by @liopeer in #343
- Resume MLflow runs when resuming from interrupted by @yutong-xiang-97 in #334
- Use
dataandoutas mmap hash for train task by @yutong-xiang-97 in #344 - Add Docs for using DINOv3 EoMT Model Weights by @yutong-xiang-97 in #348
- Add Custom Mask Names to Docs by @yutong-xiang-97 in #349
- Remove Upper Limit for PyTorch Version by @yutong-xiang-97 in #350
- Add class mapping support to docs by @yutong-xiang-97 in #355
- Infer image shape from model for onnx export by @simonschoelly in #307
- Add Docs for Multi-channel Image Support by @yutong-xiang-97 in #356
- Temporarily disable flaky test by @simonschoelly in #358
- Add RandomIoUCrop by @liopeer in #320
- Update README.md by @matthiasheller in #330
- Update README with auto-labeling results by @yutong-xiang-97 in #359
- Replace types-pillow with pillow by @simonschoelly in #361
- Add docs for checkpoints, exported models, and resume training by @yutong-xiang-97 in #357
- Fix failing export tests by @stegmuel in #362
- Add Banner to Docs by @liopeer in #365
- Fix Torch Version in CI by @yutong-xiang-97 in #368
- enable p!=1.0 by @liopeer in #367
- Add instance segmentation dataset with yolo format by @guarin in #369
- Fix NormalizeArgs from dict by @guarin in #374
- Fancy-Up the CSS by @liopeer in #366
- Download checkpoint if it does not exist as a file by @simonschoelly in #373
- Fix loading model weights from model init args by @yutong-xiang-97 in #372
- Add eomt instance segmentation transform by @guarin in #378
- Fix Avg FPS in README for COCO stuff by @liopeer in #386
- Add EoMT instance segmentation loss by @guarin in #382
- Update exclude newer by @guarin in #383
- Implement predict semantic segmentation with
model.predictby @yutong-xiang-97 in #384 - Fix same mmapfile by @stegmuel in #387
- Fix CI errors: add
deviceparameter to the predict function by @yutong-xiang-97 in #393 - Fix local mypy errors by @guarin in #394
- Document predict semantic segmentation by @yutong-xiang-97 in #391
- Add citation.cff file by @stegmuel in #395
- Add vitl16-ade20k checkpoint by @yutong-xiang-97 in #398
- Remove need for DINOv3 URL and add ConvNeXt Distillation by @liopeer in #380
- Reduce memory usage for predict and update model hash by @yutong-xiang-97 in #399
- Update Semantic Segmentation Notebook by @guarin in #401
- Update DINOv3 docs by @guarin in #400
- EoMT instance segmentation task model by @guarin in #389
- Implement Object Detection Training by @liopeer in #364
- Thomas trn 1627 randomioucrop correctness by @stegmuel in #407
- Update contributing guide by @yutong-xiang-97 in #405
- Add EoMT instance segmentation train model by @guarin in #404
- EoMT instance segmentation handle image size by @guarin in #408
- Thomas trn 1612 test od by @stegmuel in #411
- Use float32 for numpy image and separate image and mask opening by @yutong-xiang-97 in #409
- Allow pretrained models as model argument by @guarin in #410
- Ignore TF32 warning by @guarin in #412
- Support URLs in predict by @guarin in #415
- Thomas trn 1541 prerelease fixes by @stegmuel in #413
- Release 0.12.0 by @guarin in #414
New Contributors
- @matthiasheller made their first contribution in #330
Full Changelog: v0.11.4...v0.12.0
v0.11.4
[0.11.4] - 2025-10-08
Added
- Add support for saving the best semantic segmentation checkpoints and model weights during training.
- Expose more arguments for the checkpointing callback in pretraining.
- Add LT-DETR inference support with DINOv2 and DINOv3 ConvNext backbones.
Changed
- Change default precision to
bf16-mixedfor pretraining on GPUs that support it. - Fix warning about too few epochs for DINOv2 which occurs with the default epoch calculation.
What's Changed
- Update readme with the benchmark results by @yutong-xiang-97 in #316
- Mention license info for DINOv3 weights in README by @yutong-xiang-97 in #319
- Update README conference banner by @liopeer in #322
- chore: add a link to the social network YouTube by @Olexandr88 in #317
- Add Obj Det Transform Pt 2 by @liopeer in #300
- Fix FPS for ViT-S on ADE20k by @yutong-xiang-97 in #328
- Support saving best model and metric with EoMT by @yutong-xiang-97 in #326
- Default to bf16-mixed for Pretraining by @liopeer in #331
- Fix model weights loading when continuing training from a model weight file by @yutong-xiang-97 in #324
- Use floor operation in determining DINOv2 epochs by @liopeer in #335
- Add Object Detection Inference by @liopeer in #325
- Expose all ModelCheckpoint Callback Arguments by @liopeer in #342
- Table with OD results by @stegmuel in #340
- Prepare Release 0.11.4 by @liopeer in #341
New Contributors
- @Olexandr88 made their first contribution in #317
Full Changelog: v0.11.3...v0.11.4
v0.11.3
[0.11.3] - 2025-09-25
Added
- Add EoMT semantic segmentation benchmark results and model weights trained on ADE20k, COCO-Stuff, and Cityscapes datasets.
- Add support for exporting the semantic segmentation model weights to
exported_models/exported_last.pt. - Add support for allow loading semantic segmentation model weights for training.
- Add
simplifyflag to ONNXexport_task. - Add support for using DINOv3 models as teacher in distillationv1.
Fixed
- Fix a bug in
model.predict()withignore_index. - Speed up listing of filenames in large datasets.
What's Changed
- Speed up file listing with list of filepaths by @guarin in #283
- Increase ci depdency cutoff date to 2025-08-07 by @simonschoelly in #282
- Add support for dinov3 teachers in distillationv1 by @stegmuel in #281
- Use image size everywhere in predict by @guarin in #288
- Memory-map file follow-up by @yutong-xiang-97 in #287
- Add onnxslim by @simonschoelly in #277
- Use flash attention for dinov2 eomt segmentation task by @simonschoelly in #286
- Fix grid size in EoMT with DINOv2 ViT + add config for ViT-L/16 for DINOv2 ViT by @stegmuel in #293
- Fix grid size in linear segmentation by @stegmuel in #295
- Publish EoMT DINOv3 Results by @yutong-xiang-97 in #291
- Update readme by @guarin in #294
- Lightly at MICCAI by @guarin in #299
- Object Detection Transforms by @liopeer in #292
- Resume scheduler properly by @LinkinPony in #301
- Unique cache dir per test by @guarin in #302
- Add fallback for onnx export when tiling is not necessary by @simonschoelly in #297
- Change ignore_index pred behavior by @yutong-xiang-97 in #305
- Improve existing mmap file handling by @guarin in #303
- Document semantic segmentation onnx export by @simonschoelly in #298
- Support loading checkpoint for semantic segmentation by @yutong-xiang-97 in #308
- Add Loading Checkpoints to Docs by @yutong-xiang-97 in #309
- Export EoMT model to
.ptat the end of training by @yutong-xiang-97 in #312 - Add Colab notebook examples for EoMT inference with the model weights by @yutong-xiang-97 in #313
- Report benchmark results and link to model weights in the docs by @yutong-xiang-97 in #314
- Release v0.11.3 by @yutong-xiang-97 in #315
New Contributors
- @LinkinPony made their first contribution in #301
Full Changelog: v0.11.2...v0.11.3