Paper - Domain Adaptive Knowledge Distillation for Driving Scene Semantic Segmentation

Please cite our paper if you find it useful.

@inproceedings{kothandaramandomain,
  title={Domain Adaptive Knowledge Distillation for Driving Scene Semantic Segmentation},
  author={Kothandaraman, Divya and Nambiar, Athira and Mittal, Anurag},
  booktitle={Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision},
  pages={134--143}
}

Table of Contents

• Paper - Domain Adaptive Knowledge Distillation for Driving Scene Semantic Segmentation
• Repo Details and Contents
  • Code structure
  • Datasets
  • Dependencies
• Our network
• Acknowledgements

Repo Details and Contents

Python version: 3.7

Code structure

Dataoaders - The dataloaders can be found in the 'dataset' folder, along with the image lists.
Models - Contains code for the network architectures
Utils - Contains code for the cross entropy loss function

Training

train_gta2cs_multi_drnd38, train_gta2cs_multi_drnd22 - Training script for teacher and undistilled student networks
train_gta2cs_ts_multi.py - Training script for the domain adaptive knowledge distillation network

Evaluation

eval_cs.py - Contains the evaluation script for cityscapes

Datasets

• CityScapes
• Berkeley Deep Drive
• GTA5

Dependencies

PyTorch
NumPy
SciPy
Matplotlib

Our network

Acknowledgements

This code is heavily borrowed from AdaptSegNet