Weakly Supervised Co-training with Swapping Assignments for Semantic Segmentation

27 Feb 2024  ·  Xinyu Yang, Hossein Rahmani, Sue Black, Bryan M. Williams ·

Class activation maps (CAMs) are commonly employed in weakly supervised semantic segmentation (WSSS) to produce pseudo-labels. Due to incomplete or excessive class activation, existing studies often resort to offline CAM refinement, introducing additional stages or proposing offline modules. This can cause optimization difficulties for single-stage methods and limit generalizability. In this study, we aim to reduce the observed CAM inconsistency and error to mitigate reliance on refinement processes. We propose an end-to-end WSSS model incorporating guided CAMs, wherein our segmentation model is trained while concurrently optimizing CAMs online. Our method, Co-training with Swapping Assignments (CoSA), leverages a dual-stream framework, where one sub-network learns from the swapped assignments generated by the other. We introduce three techniques: i) soft perplexity-based regularization to penalize uncertain regions; ii) a threshold-searching approach to dynamically revise the confidence threshold; and iii) contrastive separation to address the coexistence problem. CoSA demonstrates exceptional performance, achieving mIoU of 76.2\% and 51.0\% on VOC and COCO validation datasets, respectively, surpassing existing baselines by a substantial margin. Notably, CoSA is the first single-stage approach to outperform all existing multi-stage methods including those with additional supervision. Code is avilable at \url{https://github.com/youshyee/CoSA}.

PDF Abstract

Datasets


Task Dataset Model Metric Name Metric Value Global Rank Benchmark
Weakly-Supervised Semantic Segmentation COCO 2014 val CoSA (ViT-B, single-stage) mIoU 51.1 # 6
Weakly-Supervised Semantic Segmentation COCO 2014 val CoSA (SWIN-B, multi-stage) mIoU 53.7 # 5
Weakly-Supervised Semantic Segmentation PASCAL VOC 2012 val CoSA (SWIN-B, multi-stage) Mean IoU 81.4 # 4
Weakly-Supervised Semantic Segmentation PASCAL VOC 2012 val CoSA (ViT-B, single-stage) Mean IoU 76.4 # 8

Methods