lightly-ai/lightly-train

View on GitHub

All-in-one training for vision models (YOLO, ViTs, RT-DETR, DINOv3): pretraining, fine-tuning, distillation.

Stars: 1.7k
Forks: 116
Open beginner issues: 0
Indexed issues: 56
Avg merge: 2d 21h
Merged PRs (30d): 6

Dominant language: Python
License: AGPL-3.0
Last GitHub push: Sep 14, 2026
Latest indexed: Sep 19, 2026
Contributing guide: Contributing guide
Code of conduct: No code of conduct
Beginner labels: help wanted good first issue

Beginner friendly Everything

Open All

56 open issues indexed

[Bug] DenseCL updates its key encoder and queues between accumulated microbatches Open

Difficulty 4/5 3-5 days Newbie friendliness 55/100

lightly-ai/lightly-train#973 · 1 comment · Opened Sep 17, 2026
Verify evals on Papers with Code Open

Difficulty 4/5 3-5 days Newbie friendliness 45/100

lightly-ai/lightly-train#971 · 1 comment · 1 reaction · Opened Sep 14, 2026
[Bug] DINO teacher EMA and center update once per microbatch with gradient accumulation Open

bug

Difficulty 4/5 3-5 days Newbie friendliness 48/100

lightly-ai/lightly-train#968 · 1 comment · Opened Sep 10, 2026
[FEAT] Support class weights in the loss for image classification Open

enhancement help wanted

Difficulty 4/5 3-5 days Newbie friendliness 68/100

lightly-ai/lightly-train#966 · 1 comment · Opened Sep 3, 2026
[FEAT] Support class weights in the loss to handle imbalanced datasets Open

enhancement tracking issue

Difficulty 5/5 Over a week Newbie friendliness 45/100

lightly-ai/lightly-train#965 · Opened Sep 3, 2026
[QUESTION] [Multi-label classification] Does an unbalanced class distribution cause problems during training? Open

question

Difficulty 5/5 Over a week Newbie friendliness 30/100

lightly-ai/lightly-train#964 · 2 comments · Opened Sep 2, 2026
Export a pretrained embedding model (dinov2/dinov3) without a training run Open

Difficulty 4/5 3-5 days Newbie friendliness 48/100

lightly-ai/lightly-train#963 · 3 comments · Opened Sep 1, 2026
[DOC] Clarify Vision Foundry listing under “LightlyTrain in Research” Open

documentation

Difficulty 2/5 1-3 hours Newbie friendliness 68/100

lightly-ai/lightly-train#957 · 2 comments · Opened Aug 25, 2026
[QUESTION] Can a multilabel model predict label combinations not seen during training? Open

question

Difficulty 4/5 3-5 days Newbie friendliness 35/100

lightly-ai/lightly-train#943 · 2 comments · Opened Aug 12, 2026
EoMT validation uses much more GPU memory than training Open

bug dinov2 dinov3 help wanted segmentation tracking issue

Difficulty 3/5 1-2 days Newbie friendliness 72/100

lightly-ai/lightly-train#941 · 3 comments · Opened Aug 12, 2026
[QUESTION] EoMT validation OOM because masks are resized to original image size Open

question

Difficulty 4/5 3-5 days Newbie friendliness 50/100

lightly-ai/lightly-train#936 · 4 comments · Opened Aug 10, 2026
[FEAT] TensorRT INT8 quantization support + TRT 11.x compatibility Open

enhancement

Difficulty 5/5 Over a week Newbie friendliness 38/100

lightly-ai/lightly-train#935 · 1 comment · Opened Aug 10, 2026
[BUG] DINOv3 LT-DETR: -notpretrained backbone aliases broken or missing for most sizes Open

bug dinov3 documentation object-detection

Difficulty 4/5 3-5 days Newbie friendliness 66/100

lightly-ai/lightly-train#933 · 1 comment · Opened Aug 7, 2026
Finalize EdgeCrafter backbones package implementation Open

Difficulty 3/5 1-2 days Newbie friendliness 48/100

lightly-ai/lightly-train#928 · 1 comment · Opened Aug 3, 2026
Add multi-scale feature support to Torchvision Open

Difficulty 4/5 3-5 days Newbie friendliness 35/100

lightly-ai/lightly-train#927 · Opened Aug 3, 2026
Add multi-scale feature support to TIMM Open

Difficulty 4/5 3-5 days Newbie friendliness 38/100

lightly-ai/lightly-train#926 · 1 comment · Opened Aug 3, 2026
Add multi-scale feature support to all backbone packages Open

distillation enhancement good first issue help wanted models pretraining tracking issue

Difficulty 4/5 3-5 days Newbie friendliness 66/100

lightly-ai/lightly-train#925 · 6 comments · Opened Aug 3, 2026
[DOC] Guide: evaluating pretraining/distillation on a custom detection dataset Open

documentation

Difficulty 4/5 3-5 days Newbie friendliness 45/100

lightly-ai/lightly-train#924 · 3 comments · Opened Aug 3, 2026
Document C++ Inference Examples Open

enhancement good first issue help wanted object-detection

lightly-ai/lightly-train#921 · 3 comments · 1 assignee · Opened Jul 30, 2026
Questions about continued pretraining DINOv2 and feature extraction in lightly_train Open

question

Difficulty 4/5 3-5 days Newbie friendliness 35/100

lightly-ai/lightly-train#821 · 13 comments · Opened Jun 28, 2026
[FEAT] FFCV implementation for dataset Open

enhancement

Difficulty 5/5 Over a week Newbie friendliness 35/100

lightly-ai/lightly-train#771 · 1 comment · Opened Jun 10, 2026
[FEAT] <Brief Description>ConvNext for semantic segmentation Open

enhancement

Difficulty 5/5 Over a week Newbie friendliness 30/100

lightly-ai/lightly-train#769 · 1 comment · Opened Jun 10, 2026
[QUESTION] <Brief Description> convnext-large-ltdetr-coco train log Open

question

Difficulty 4/5 3-5 days Newbie friendliness 35/100

lightly-ai/lightly-train#766 · 4 comments · Opened Jun 6, 2026
[FEAT] DEIMv2 implementation Open

enhancement

Difficulty 5/5 Over a week Newbie friendliness 45/100

lightly-ai/lightly-train#756 · 1 comment · Opened May 28, 2026
[BUG] Model Training Crashing Windows Open

bug

Difficulty 4/5 3-5 days Newbie friendliness 38/100

lightly-ai/lightly-train#703 · 1 comment · Opened Apr 15, 2026
[FEAT] Add DINOV3 pre-training Open

enhancement

Difficulty 5/5 Over a week Newbie friendliness 28/100

lightly-ai/lightly-train#692 · 4 comments · Opened Apr 8, 2026
[FEAT] JEPA framework Open

enhancement

Difficulty 5/5 Over a week Newbie friendliness 25/100

lightly-ai/lightly-train#685 · 2 comments · Opened Apr 5, 2026
[QUESTION] License information Open

question

Difficulty 1/5 Under an hour Newbie friendliness 25/100

lightly-ai/lightly-train#671 · 1 comment · Opened Mar 26, 2026
[DOC] Can we use a bigger Dinov3 model? Open

documentation

Difficulty 4/5 3-5 days Newbie friendliness 35/100

lightly-ai/lightly-train#637 · 1 comment · Opened Mar 9, 2026
[FEAT] <Brief Description> Open

enhancement

Difficulty 5/5 Over a week Newbie friendliness 25/100

lightly-ai/lightly-train#634 · 1 comment · Opened Mar 6, 2026
[QUESTION] distill dinov3 to repvit Open

question

Difficulty 5/5 Over a week Newbie friendliness 25/100

lightly-ai/lightly-train#628 · 1 comment · Opened Feb 28, 2026
Why is there no confidence threshold for dinov3_eomt_semantic_segmentation predict Open

Difficulty 3/5 1-2 days Newbie friendliness 45/100

lightly-ai/lightly-train#583 · 6 comments · Opened Jan 26, 2026
[QUESTION] Pretraining or Using Huggingface models Open

question

Difficulty 5/5 Over a week Newbie friendliness 25/100

lightly-ai/lightly-train#570 · 8 comments · Opened Jan 16, 2026
WebDataset Open

enhancement

Difficulty 4/5 3-5 days Newbie friendliness 25/100

lightly-ai/lightly-train#564 · 3 comments · Opened Jan 15, 2026
Early Stopping Open

documentation

Difficulty 4/5 3-5 days Newbie friendliness 25/100

lightly-ai/lightly-train#549 · 1 comment · Opened Jan 11, 2026
DinoV3 Support Open

enhancement

Difficulty 5/5 Over a week Newbie friendliness 25/100

lightly-ai/lightly-train#548 · 3 comments · Opened Jan 11, 2026
[QUESTION] Distillation Pretraining DINOv3 Open

question

Difficulty 3/5 1-2 days Newbie friendliness 35/100

lightly-ai/lightly-train#544 · 5 comments · Opened Jan 9, 2026
[QUESTION] Custom albumentations for Detection / Instance Segmentation Open

question

Difficulty 5/5 Over a week Newbie friendliness 25/100

lightly-ai/lightly-train#530 · 6 comments · Opened Jan 2, 2026
[FEAT] Support for OBB detection Open

enhancement

Difficulty 5/5 Over a week Newbie friendliness 25/100

lightly-ai/lightly-train#510 · 12 comments · 9 reactions · Opened Dec 16, 2025
Custom distillation model using DinoVol3 does not achieve the required accuracy after int8 quantization Open

Difficulty 4/5 3-5 days Newbie friendliness 35/100

lightly-ai/lightly-train#500 · 3 comments · Opened Dec 11, 2025
[FEAT] Integrating Sam3 Open

enhancement

Difficulty 5/5 Over a week Newbie friendliness 25/100

lightly-ai/lightly-train#491 · 3 comments · Opened Dec 10, 2025
[QUESTION] Paper describing LT-DETR Open

question

Difficulty 1/5 Under an hour Newbie friendliness 20/100

lightly-ai/lightly-train#470 · 3 comments · Opened Nov 25, 2025
[QUESTION] load support Open

question

Difficulty 5/5 Over a week Newbie friendliness 25/100

lightly-ai/lightly-train#459 · 2 reactions · Opened Nov 20, 2025
[FEAT] Support for DINOv3 Self-Supervised Training Open

enhancement

Difficulty 5/5 Over a week Newbie friendliness 30/100

lightly-ai/lightly-train#457 · 5 comments · 8 reactions · Opened Nov 20, 2025
[QUESTION] <Image and mask paths are read from a text file for segmentation model?> Open

question segmentation

Difficulty 3/5 1-2 days Newbie friendliness 35/100

lightly-ai/lightly-train#443 · 4 comments · Opened Nov 14, 2025
How to perform self-distillation of DINOv3 using unlabeled data？ Open

dinov3 question

Difficulty 5/5 Over a week Newbie friendliness 20/100

lightly-ai/lightly-train#436 · 3 comments · Opened Nov 13, 2025
[BUG] <Pre-training weight loading issue> Open

bug dinov3 segmentation

Difficulty 3/5 1-2 days Newbie friendliness 42/100

lightly-ai/lightly-train#435 · 9 comments · 1 reaction · Opened Nov 13, 2025
[QUESTION] Handling Unlabeled and Background Classes in Mask2Former for Custom Dataset Open

medical imaging question segmentation

Difficulty 5/5 Over a week Newbie friendliness 25/100

lightly-ai/lightly-train#403 · 5 comments · 1 reaction · Opened Nov 3, 2025
[FEAT] Add Attention Visualization for Vision Transformers (ViT/DINOv2) and ConvNets Open

enhancement

Difficulty 5/5 Over a week Newbie friendliness 35/100

lightly-ai/lightly-train#333 · 5 comments · 1 reaction · Opened Oct 1, 2025
[FEAT] <Brief Description>When will the training of Dinov3 be supported Open

enhancement

Difficulty 5/5 Over a week Newbie friendliness 35/100

lightly-ai/lightly-train#304 · 8 comments · Opened Sep 18, 2025

lightly-ai/lightly-train

Get new issues in your inbox