BoMuDA: Boundless Multi-Source Domain Adaptive Segmentation in Unconstrained Environments

Overview of BoMuDA:The input consists of N sources from which the Best-Source is selected by the Alt-Inc algorithm. The Alt-Inc algorithm proceeds in an unsupervised fashion to generate the final set of pseudo-labels that are used to perform boundless DA. The final output consists of the segmentation map of an image in the target domain.

Paper Code Supplementary Materials
BoMuDA GitHub Code Coming Soon

We present an unsupervised multi-source domain adaptive semantic segmentation approach in unstructured and unconstrained traffic environments. We propose a novel training strategy that alternates between single-source domain adaptation (DA) and multi-source distillation, and also between setting up an improvised cost function and optimizing it. In each iteration, the single-source DA first learns a neural network on a selected source, which is followed by a multi-source fine-tuning step using the remaining sources. We call this training routine the Alternating-Incremental ("Alt-Inc") algorithm. Furthermore, our approach is also boundless i.e. it can explicitly classify categories that do not belong to the training dataset (as opposed to labeling such objects as "unknown"). We have conducted extensive experiments and ablation studies using the Indian Driving Dataset, CityScapes, Berkeley DeepDrive, GTA V, and the Synscapes datasets, and we show that our unsupervised approach outperforms other unsupervised and semi-supervised SOTA benchmarks by 5.17% - 42.9% with a reduced model size by up to 5.2x.

Please cite our work if you found it useful,

  title={BoMuDA: Boundless Multi-Source Domain Adaptive Segmentation in Unconstrained Environments},
  author={Kothandaraman, Divya and Chandra, Rohan and Manocha, Dinesh},
  journal={arXiv preprint arXiv:2010.03523},