Efficient-Segmentation-Networks

Why we included this project

Choosing a lightweight segmentation network usually means juggling a dozen separate repos, each with its own quirks. This one collects many of the efficient encoder-decoder models people actually use for real-time work, from ENet and ERFNet to ESPNet, LEDNet, CGNet and Fast-SCNN, all written in a consistent PyTorch style. That consistency matters because you can compare backbones and swap them without relearning a new codebase each time. The code is set up around Cityscapes and CamVid, the standard driving-scene benchmarks, and bundles the loss functions, optimizers, normalization tricks, and learning-rate schedules you need to reproduce or adapt training runs. It is a reference toolbox rather than a polished product, so expect to read the code and fit it to your own pipelines, which suits researchers, students, and engineers prototyping segmentation on embedded or low-latency systems.