Multi-scale and Cross-scale Contrastive Learning for Semantic Segmentation

TL;DR

This paper introduces a novel multi-scale and cross-scale contrastive learning framework for semantic segmentation that enhances feature discriminability without requiring data augmentation or memory banks, improving performance across diverse models and datasets.

Contribution

It proposes a new contrastive learning method leveraging multi-scale features and cross-scale constraints, applicable to various models and datasets, without needing additional data augmentation or memory modules.

Findings

Boosts performance of models like DeepLabV3, HRNet, OCRNet, UPerNet

Effective across natural and surgical datasets

No need for data augmentation or online memory banks

Abstract

This work considers supervised contrastive learning for semantic segmentation. We apply contrastive learning to enhance the discriminative power of the multi-scale features extracted by semantic segmentation networks. Our key methodological insight is to leverage samples from the feature spaces emanating from multiple stages of a model's encoder itself requiring neither data augmentation nor online memory banks to obtain a diverse set of samples. To allow for such an extension we introduce an efficient and effective sampling process, that enables applying contrastive losses over the encoder's features at multiple scales. Furthermore, by first mapping the encoder's multi-scale representations to a common feature space, we instantiate a novel form of supervised local-global constraint by introducing cross-scale contrastive learning linking high-resolution local features to low-resolution global features. Combined, our multi-scale and cross-scale contrastive losses boost performance of various models (DeepLabV3, HRNet, OCRNet, UPerNet) with both CNN and Transformer backbones, when evaluated on 4 diverse datasets from natural (Cityscapes, PascalContext, ADE20K) but also surgical (CaDIS) domains. Our code is available at https://github.com/RViMLab/MS_CS_ContrSeg. datasets from natural (Cityscapes, PascalContext, ADE20K) but also surgical (CaDIS) domains.