CA-SSL: Class-Agnostic Semi-Supervised Learning for Detection and Segmentation

TL;DR

This paper introduces CA-SSL, a class-agnostic semi-supervised learning framework that balances task-specific training signals to improve detection and segmentation performance, avoiding underfitting and overfitting issues.

Contribution

The paper proposes a novel CA-SSL framework with a warmup stage that ignores class info in pseudo labels, enhancing detection and segmentation results over traditional SSL methods.

Findings

Achieves 4.7% performance gain over baseline on FCOS detection.

Warmup model better avoids underfitting/overfitting in downstream tasks.

Demonstrates strong transferability to other detection and segmentation frameworks.

Abstract

To improve instance-level detection/segmentation performance, existing self-supervised and semi-supervised methods extract either task-unrelated or task-specific training signals from unlabeled data. We show that these two approaches, at the two extreme ends of the task-specificity spectrum, are suboptimal for the task performance. Utilizing too little task-specific training signals causes underfitting to the ground-truth labels of downstream tasks, while the opposite causes overfitting to the ground-truth labels. To this end, we propose a novel Class-Agnostic Semi-Supervised Learning (CA-SSL) framework to achieve a more favorable task-specificity balance in extracting training signals from unlabeled data. CA-SSL has three training stages that act on either ground-truth labels (labeled data) or pseudo labels (unlabeled data). This decoupling strategy avoids the complicated scheme in traditional SSL methods that balances the contributions from both data types. Especially, we introduce a warmup training stage to achieve a more optimal balance in task specificity by ignoring class information in the pseudo labels, while preserving localization training signals. As a result, our warmup model can better avoid underfitting/overfitting when fine-tuned on the ground-truth labels in detection and segmentation tasks. Using 3.6M unlabeled data, we achieve a significant performance gain of 4.7% over ImageNet-pretrained baseline on FCOS object detection. In addition, our warmup model demonstrates excellent transferability to other detection and segmentation frameworks.