FORECASS: Forgetting-Resilient Continuous Source-Free UDA for Semantic Segmentation

Abstract

Semantic segmentation is fundamental for autonomous driving, requiring dense scene understanding across diverse and continu- ously changing environments. Real-world deployment must overcome two major challenges: (1) dynamic environmental shifts due to weather, lighting, and geography, and (2) the inability to retain labeled source data for continual adaptation. To address these issues, we propose a novel source-free Adaptive and Forgetting-Resilient Continual Unsupervised Domain Adaptation for Semantic Segmentation (FORECASS). We introduce a teacher–student based framework, using EMA (Exponential Moving Average) updat- ing technique, to produce stable pseudo-labels during continual adaptation. Central to our method is a refiner-based error estimation model that predicts pixel-wise pseudo-label reliability during adaptation. By leveraging the error map, the model selectively fo- cuses learning on uncertain and challenging regions, playing a critical role in mitigating catastrophic forgetting. Complementarily, a structure-aware consistency mechanism enforces semantic coherence across views, further enhancing stability during sequential adaptation. Lightweight knowledge distillation is also incorporated to smooth alignment between the pseudo-label generator and the adaptation model. We validate our framework on the challenging continual adaptation sequence GTA → Cityscapes → IDD → Mapillary, achieving state-of-the-art results over both source-dependent and source-free baselines.

Offline evaluation (mIoU)

This repository provides offline_validation.py, a standalone evaluation script for computing semantic segmentation quality (mean IoU) for FORECASS. Full training and adaptation code for FORECASS will be available soon (after acceptance of paper).

Data

A download link https://utbm-my.sharepoint.com/:f:/g/personal/baqar_abbas_utbm_fr/EorrpuM9biZGhneKXxDDdccBHZsnkvyqHaMHYWgVwJcSYw?e=fFAUGL is provided with:

• Predicted semantic segmentation maps generated by FORECASS.
• Ground-truth label maps for two taxonomies:
  • SOTA mapping (baseline 19→7 consolidation used in prior work).
  • FORECASS revised 7-class mapping.

Both sets follow the same filename pattern:

• <id>_leftImg8bit.png for predictions
• <id>_gtFine_labelIds.png for ground truth
  where <id> may be e.g. 000290, frankfurt_000000_000294, mapillary_000000.
  The script supports both flat folders and Cityscapes-style subfolders.

Usage

1. Download and unzip the provided prediction and GT folders.
2. Edit the paths at the top of offline_validation.py:

   PRED_ROOT = r"PATH_TO_PRED_FOLDER"
   GT_ROOT   = r"PATH_TO_GT_FOLDER"

Run

python offline_validation.py