.png&w=640&q=75)
PidNet: Semantic Segmentation
PidNet, proposed in 2023, is a real-time semantic segmentation model that innovatively integrates Proportional-Integral-Derivative (PID) control theory into feature fusion. Its architecture consists of three parallel branches: a Proportional Branch (capturing local details), an Integral Branch (aggregating global context), and a Derivative Branch (detecting edge variations). These are adaptively fused via a PID controller, dynamically balancing multi-scale features for accuracy and speed. With a lightweight design, PidNet achieves 80%+ mIoU on Cityscapes at over 100 FPS, outperforming models like BiSeNet. It excels in segmenting small objects and fine edges in complex scenes, making it ideal for autonomous driving, robotics, and industrial inspection where real-time processing and precision are critical.
Source model
- Input shape: 1x3x1024x2048
- Number of parameters: 7.62M
- Model size: 29.14M
- Output shape: 1x19x128x256
The source model can be found here
Performance Reference
Please search model by model name in Model Farm
Inference & Model Conversion
Please search model by model name in Model Farm
License
Source Model: MIT
Deployable Model: APLUX-MODEL-FARM-LICENSE