PaddleSeg

PaddleSeg is an image segmentation toolkit for developers building computer vision applications on their own hardware. It covers annotation, training and deployment for tasks such as medical image analysis, industrial inspection and portrait effects.

• Scene and object segmentation: Label image regions by category with models such as DeepLabV3, OCRNet and SegFormer, or use panoptic segmentation to distinguish individual objects alongside scene categories.
• Interactive annotation and matting: Guide image selections with EISeg and separate subjects from backgrounds with matting models such as PP-Matting and MODNet.
• Specialized segmentation: Build models for 3D medical images, portrait segmentation and multi-label tasks where a region can have more than one label.
• Training and compression: Adapt pretrained models to your datasets, train across multiple GPUs, and use quantization, distillation or pruning to reduce model size and inference cost.
• Deployment and integration: Deploy through Paddle Inference, Paddle Lite or Paddle Serving, export ONNX models, or use PaddleX for model development through Python APIs and a graphical interface.

PaddleSeg is open source under Apache 2.0 and uses the PaddlePaddle framework. Hardware needs depend on the model: heavier models target server GPUs and Jetson devices, while lighter choices such as PP-LiteSeg and PP-MobileSeg can run on x86 server CPUs and mobile ARM CPUs. NVIDIA GPU inference supports TensorRT acceleration.