Instance Segmentation

LightlyTrain supports fine-tuning instance segmentation models with the train_instance_segmentation function. Two model families are currently supported:

LTDETRv2 (NEW)

Open In Colab

LightlyTrain’s LTDETRv2 is a DETR-based instance segmentation family built on ECViT backbones from the EdgeCrafter paper by Liu et al. It significantly outperforms YOLO11 and achieves the best accuracy-vs-parameter-count trade-off among current SOTA instance segmentation models! See LTDETRv2 for benchmark results, available models, and how to train, predict, and export a LTDETRv2 instance segmentation model.

EoMT

Open In Colab

LightlyTrain also supports training DINOv2 and DINOv3-based instance segmentation models with the EoMT architecture by Kerssies et al. See EoMT for benchmark results, available models, and how to train, predict, and export an EoMT instance segmentation model.