HagSeg: Hardness-adaptive Guidance for Semi-supervised Semantic Segmentation

Zhen Zhao, Tianyi Wu, Jimin Pi, Sifan Long, Luping Zhou, Jingdong Wang

Keywords: Semi-supervised, Semantic Segmentation, Hardness-adaptive guidance

Abstract: Recently, semi-supervised semantic segmentation has achieved promising performance with a small fraction of labelled data. However, most existing studies treat all unlabeled data equally and barely consider the differences and training difficulties among unlabeled instances. Differentiating unlabeled instances can promote instance-specific supervision to adapt to the model's evolution dynamically. In this paper, we emphasize the cruciality of instance differences and propose an instance-specific and hardness-adaptive guidance for semi-supervised semantic segmentation, named as HagSeg. Relying on the model's performance, HagSeg employs the class-weighted symmetric intersection-over-union to evaluate the hardness of each unlabeled instance and then supervises the training on unlabeled data in a hardness-adaptive manner. Specifically, HagSeg learns from unlabeled instances progressively by weighing their corresponding consistency losses based on the evaluated hardness. Meanwhile, HagSeg dynamically adjusts the augmentation for each instance such that the distortion degree of augmented instances is adapted to the model's generalization capability across the training course. Not integrating additional losses and training procedures, HagSeg can obtain remarkable performance gains against current state-of-the-art approaches on segmentation benchmarks under different semi-supervised partition protocols.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: General Machine Learning (ie none of the above)

TL;DR: A instance-specific and hardness-adaptive SSS framework