Releases · lightly-ai/lightly-train

11 min read Original article ↗

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

Full Changelog: v0.13.1...v0.13.2

v0.13.1

Fixed

  • Fix bug in ONNX export for object detection models.

What's Changed

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

Changed

  • Rename lightly_train.train() to lightly_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

New Contributors

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_head argument in train command. 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_size not tuple when training from pretrained model.
  • Fix a bug when fine-tuning a model with resume_interrupted=True.
  • Fix num_classes not updated when loading an object detection checkpoint with
    different number of classes.

What's Changed

Full Changelog: v0.12.2...v0.12.3

v0.12.2

Fixed

  • Fix image_size not tuple when training from pretrained model.

What's Changed

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

New Contributors

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 model argument
    to the model training functions like train_semantic_segmentation.

Changed

  • Widen PyTorch constraint — remove <2.6 upper bound to allow PyTorch 2.6 and later that is officially supported by PyTorch Lightning 2.5.
  • Rename load_model_from_checkpoint to load_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

New Contributors

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-mixed for pretraining on GPUs that support it.
  • Fix warning about too few epochs for DINOv2 which occurs with the default epoch calculation.

What's Changed

New Contributors

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 simplify flag to ONNX export_task.
  • Add support for using DINOv3 models as teacher in distillationv1.

Fixed

  • Fix a bug in model.predict() with ignore_index.
  • Speed up listing of filenames in large datasets.

What's Changed

New Contributors

Full Changelog: v0.11.2...v0.11.3