This library provides trainable semantic segmentation models implemented in C++ using LibTorch (PyTorch C++ frontend). It includes popular segmentation architectures such as U-Net, FPN, PAN, LinkNet, PSPNet, DeepLab-V3, and DeepLab-V3+. The library supports multiple encoder backbones including ResNet, VGG, and ResNeXt, all with pretrained ImageNet weights. Users can create and deploy segmentation models with a simple high-level API and benefit from faster inference compared to Python PyTorch.