Multi-level Collaborative Learning for Multi-Target Domain Adaptive Semantic Segmentation

Datasets

By default, the datasets are put in <root_dir>/Dataset. Similar to CoaST.

step 1: Pre-trained models

sh pretrain.sh

step 2: Pseudo-labels initialization and train models

sh train.sh

step 3: Evaluation

sh test.sh

Acknowledgements

This codebase is heavily borrowed from CoaST and DAFormer. It is our pleasure to acknowledge their contributions.