The following model implementations include adaptations of Apache-2.0 source.
The license is included in Apache-2.0.txt in this directory.

MedNeXt (mednext_2d.py)
https://github.com/MIC-DKFZ/MedNeXt
Copyright 2019 Division of Medical Image Computing, German Cancer Research
Center (DKFZ), Heidelberg, Germany.

EfficientViT segmentation head (efficientvit_seg.py)
https://github.com/mit-han-lab/efficientvit
Copyright 2023 Han Cai.

IS-Net DIS segmentation network (isnet.py)
https://github.com/xuebinqin/DIS
Xuebin Qin, Hang Dai, Xiaobin Hu, Deng-Ping Fan, Ling Shao, and Luc Van Gool.
Highly Accurate Dichotomous Image Segmentation, ECCV 2022.
Source: IS-Net/models/isnet.py at b6764e20381f6f42a70f83fa3324181529ed1403.
The upstream README licenses the code and evaluation metric under Apache-2.0.
Changes: shared RSU blocks, input validation, optional input centering, and
raw logits with bicbioseg auxiliary outputs. The separate ground-truth encoder
and intermediate feature supervision training procedure are not included.
No pretrained checkpoints or DIS5K dataset files are distributed here.

DINOv3 and EfficientViT encoders are loaded through the optional timm dependency.
Pretrained encoder checkpoints are not distributed with bicbioseg. DINOv3
checkpoints retain the DINOv3 license:
https://github.com/facebookresearch/dinov3/blob/main/LICENSE.md
