Towards Source-free Domain Adaptive Semantic Segmentation via Importance-aware and Prototype-contrast Learning
Code for our paper "Towards Source-free Domain Adaptive Semantic Segmentation via Importance-aware and Prototype-contrast Learning" (Accepted by IEEE Transactions on Intelligent Vehicles)
In this paper, we propose an end-to-end source-free domain adaptation semantic segmentation method via Importance-Aware and Prototype-Contrast (IAPC) learning. The proposed IAPC framework effectively extracts domain-invariant knowledge from the well-trained source model and learns domain-specific knowledge from the unlabeled target domain.
The code is implemented with Python(3.6) and Pytorch(1.7).
- Download Cityscapes datasets
python generate_plabel_cityscapes.py
python train.py
python evaluate_cityscapes.py
Model | mIoU | mIoU* | |
---|---|---|---|
GTA5-to-Cityscapes | Source Only | 36.6 | - |
IAPC | 49.4 | - | |
Synthia-to-Cityscapes | Source Only | 35.2 | 40.5 |
IAPC | 45.3 | 52.6 |
please refer to the subdirectory OD