APD

Adaptive Perspective Distillation for Semantic Segmentation

Zhuotao Tian*; Pengguang Chen*; Xin Lai; Li Jiang; Shu Liu; Hengshuang Zhao; Bei Yu; Ming-Chang Yang; Jiaya Jia

This project provides an implementation for the TPAMI 2022 paper "Adaptive Perspective Distillation for Semantic Segmentation"

Environment

We verify our code on

• 4x3090 GPUs
• CUDA 11.1
• python 3.9
• torch 1.12.1
• torchvision 0.13.1

Other similar environments should also work properly.

Installation

git clone https://github.com/dvlab-research/APD.git
cd APD/

Results

Dataset	Student	Teacher	Baseline	Ours
ade20k	PSPNet-R18	PSPNet-R101	37.19	39.25
cityscapes	PSPNet-R18	PSPNet-R101	74.15	75.68
pascal context	PSPNet-R18	PSPNet-R101	42.29	43.96

Training

Use the following command to train PSPNet-R18 on ade20k with APD

bash ./tool/train.sh ade20k release_psp18_psp101

Citation

Please consider citing us in your publications if it helps your research.

@article{APD,
  author = {Zhuotao Tian and
            Pengguang Chen and
            Xin Lai and
            Li Jiang and
            Shu Liu and
            Hengshuang Zhao and
            Bei Yu and
            Ming{-}Chang Yang and
            Jiaya Jia},
  title = {Adaptive Perspective Distillation for Semantic Segmentation},
  journal = {{IEEE} Trans. Pattern Anal. Mach. Intell.},
  pages = {1372--1387},
  year = {2023}
}}