This repository provides segmentation models with pretrained backbones for Keras and TensorFlow Keras.
Open source
24 alternativessegmentation models.pytorch
This repository provides semantic segmentation models with more than 500 pretrained convolutional and transformer backbones.
About segmentation models.pytorch
segmentation models.pytorch is an open-source Python library for semantic image segmentation, built on PyTorch. It provides a high-level API for creating segmentation neural networks, with 12 encoder-decoder architectures such as Unet, Unet++, Segformer and DPT. The README states that the library includes more than 800 pretrained convolutional and transformer-based encoders, with timm support. The directory description gives a lower figure of more than 500 backbones.
The library also includes common training metrics and losses, such as Dice, Jaccard and Tversky. Models can be exported to ONNX and are described as compatible with TorchScript, trace and compile. Preparing input data the same way as the encoder's pretraining may improve results, but the README says this step is not required when the whole model is trained rather than only the decoder.
Key features
- Model creation with a high-level API in about two lines of code
- 12 encoder-decoder architectures, including Unet, Unet++, Segformer and DPT
- 800+ pretrained convolution- and transformer-based encoders with timm support
- Metrics and losses for training, including Dice, Jaccard and Tversky
- ONNX export of trained models
- Compatibility with TorchScript, trace and compile
Best for: Machine learning engineers and researchers who build image semantic segmentation models in PyTorch and want a choice of pretrained encoders.
Good for
- Open source, according to the directory listing.
- The README states that pretrained encoders can help training converge faster and more stably.
- The encoder range includes lightweight options for low-latency or edge inference and high-capacity options for complex tasks with many classes.
segmentation models.pytorch at a glance
- Category
- AI & Machine Learning › AI Data & Analytics
- Type
- Open-source project
- Pricing
- Not confirmed. Check the site
- Open source
- Yes · Source code
- Language
- English
- Last checked
- Oct 6, 2026 · Online when checked
- Listed
- Oct 5, 2026
segmentation models.pytorch alternatives
Tools and sites you could use instead, ranked by how closely they match.
Ultralytics YOLOv5 supports object detection, instance segmentation, classification, training, and export in PyTorch.
Open source
24 alternatives
This repository implements YOLOv3 variants in PyTorch for real time object detection, training, validation, inference, and export.
Open source
24 alternatives
The lightly Python library supports self supervised learning on images.
Open source
24 alternatives
Ultralytics supports detection, segmentation, image classification, pose estimation, and tracking across YOLO model versions.
Open source
24 alternatives
Detectron2 provides computer vision tools for object detection, segmentation, and other visual recognition tasks.
Open source
24 alternativesPopular in AI Data & Analytics
Well-known listings from the same topic.
Dotgetdot.aiForvioforvio.comFluidAI Medicalfluidai.mdApache Zeppelinzeppelin.apache.orgSola Securitysola.securityQuadraticquadratichq.com
segmentation models.pytorch FAQ
How is a segmentation model created with this library?
The README shows creating a model with smp.Unet, setting the encoder name, encoder weights, number of input channels and number of output classes.
Is data preprocessing required before training?
Not always. Preparing data the same way as during encoder pretraining may give better results, but the README says it is not necessary when training the whole model rather than only the decoder.
Can trained models be exported to ONNX?
The README lists ONNX export as a feature and includes a notebook example for exporting a trained model to ONNX.
Is something wrong or out of date?
Tell us and an editor will check it.
Own this website?
Claim this listing
Claim the listing and suggest updates yourself. An editor checks every claim first.