Verified project record

qubvel-org/segmentation_models.pytorch

segmentation_models.pytorch is a Python library that provides a high-level API for building PyTorch image semantic segmentation neural networks. It combines 12 encoder-decoder architectures with hundreds of pretrained encoders, alongside training metrics, loss functions, and model export support.

Project overview

The library lowers the barrier to building segmentation pipelines by packaging a documented two-line model creation API, broad architecture and encoder coverage, and integrated training and export tooling in a single library.

Project type: AI Coding · Image & Vision
Use cases: Coding & Development
Deployment: Refer to project documentation
License: MIT

Best for

Developers who need reusable PyTorch neural-network models for image semantic segmentation.
Developers who want a simple high-level API for combining architectures and pretrained encoders.

Key capabilities

Creates image semantic segmentation neural networks through a high-level API documented as requiring only two lines.
Provides 12 encoder-decoder segmentation architectures, including Unet, Unet++, Segformer, and DPT.
Provides hundreds of pretrained convolutional and transformer-based encoders for constructing segmentation models.
Provides popular metrics and loss functions for training routines, including Dice, Jaccard, and Tversky.
Supports ONNX export and compatibility with PyTorch scripting, tracing, and compilation workflows.

Limitations and risks

The supplied description reports 500+ pretrained backbones, while the README reports 800+ pretrained encoders. Verify the available encoder count before relying on a specific number.

Getting started

The README describes a simple high-level API and says a neural network can be created in two lines. Use the high-level Python API to create a semantic segmentation neural network.

Evidence and sources

GitHub project description: Semantic segmentation models with 500+ pretrained convolutional and transformer-based backbones.
README: **Python library with Neural Networks for Image Semantic Segmentation based on [PyTorch](https://pytorch.org/).**
README: - Super simple high-level API (just two lines to create a neural network)
README: - 12 encoder-decoder model architectures (Unet, Unet++, Segformer, DPT, ...)
README: - 800+ **pretrained** convolution- and transform-based encoders, including [timm](https://github.com/huggingface/pytorch-image-models) support

Reviewed at: 2026-07-14T23:37:19Z