qubvel-org/segmentation_models.pytorch
Segmentation Models.pytorch
This is a PyTorch semantic segmentation library designed for building image masking frameworks. It provides a collection of over 500 pretrained convolutional and transformer-based encoders and various decoder architectures to perform binary and multiclass pixel-level classification.
The library features a modular backbone integration that decouples encoder choice from decoder logic. It supports custom input channel configurations and encoder depth tuning, allowing the modification of input layers to accept non-standard channel counts while preserving pretrained weights. Some configurations also allow for the attachment of auxiliary classification heads to produce both a segmentation mask and a global image label.
Additional capabilities include preprocessing functions aligned with pretrained encoder weights and tools for exporting trained models to the ONNX format for cross-platform deployment. The system also supports integration with model hubs for saving and loading weights.
Features
- PyTorch Semantic Segmentation Libraries - Provides a comprehensive library for building image segmentation models using PyTorch with a vast collection of pretrained encoders.
- Semantic Segmentation - Provides a comprehensive library for building semantic segmentation models using pretrained encoders and various decoders.
- Binary Segmentations - Enables identifying a single object class from the background by assigning a binary label to each pixel.
- Multiclass Segmentations - Provides the capability to categorize every pixel in an image into one of several distinct object classes.
- Encoder-Decoder Architectures - Implements a variety of encoder-decoder architectures to map latent representations to pixel-level masks.
- Modular Backbone Architectures - Features a modular backbone integration that allows swapping between convolutional and transformer-based encoders.
- Pretrained Encoder Collections - Offers a collection of over 500 pretrained convolutional and transformer-based encoders for feature extraction.
- Image Masking Frameworks - Implements a system for performing binary and multiclass pixel-level classification to create object masks.
- Computer Vision Workflows - Facilitates custom vision workflows by allowing adjustments to input channels and encoder depths.
- ONNX Model Exporters - Includes tools for serializing PyTorch computational graphs into the universal ONNX format for cross-platform deployment.
- Data Preprocessing - Ships preprocessing functions that normalize and transform input images to align with pretrained encoder weights.
- Model Depth Optimization - Enables adjusting the number of downsampling operations in the encoder to balance model complexity and inference speed.
- Preprocessing Pipelines - Provides preprocessing functions aligned with specific pretrained encoder weights to ensure consistent normalization.
- Adaptive Input Layers - Allows modifying the first layer of pretrained encoders to accept custom input channel counts while preserving weights.
- Input Channel Adaptations - Allows configuring the number of input channels to support tensors with arbitrary dimensions while preserving pretrained weights.
