This is the Pytorch implementation of our work "Multi-Domain Incremental Learning for Semantic Segmentation", accepted at WACV 2022 (Algorithms Track)
Full paper: https://arxiv.org/abs/2110.12205
[New] Model checkpoints and evaluation notebook now out for easy reproducibility!
- Python 3.6
- Pytorch: Make sure to install the Pytorch version for Python 3.6 with CUDA support (This code has been run with CUDA 10.2)
- Models can take upto 40 hours on 2 Nvidia GeForce GTX 1080 Ti GPUs for Step 2; and upto 90 hours on 4 Nvidia GeForce GTX 1080 Ti GPUs for Step 3.
- Cityscapes: https://www.cityscapes-dataset.com/
- BDD100k: https://www.bdd100k.com/
- IDD: Download IDD Part 1 from https://idd.insaan.iiit.ac.in/
Preprocessing IDD: convert polygon labels to segmentation masks: