MixReorg: Cross-Modal Mixed Patch Reorganization is a Good Mask Learner for Open-World Semantic Segmentation

TL;DR

MixReorg is a novel pre-training method that improves open-world semantic segmentation by enhancing pixel-level semantic alignment through patch mixing and contrastive learning, enabling zero-shot object segmentation.

Contribution

The paper introduces MixReorg, a simple yet effective patch reorganization pre-training paradigm that significantly boosts zero-shot semantic segmentation performance without fine-tuning.

Findings

Outperforms GroupViT by 2.5-6.2% mIoU on various benchmarks.

Enables direct application to segment arbitrary categories in a zero-shot manner.

Enhances pixel-semantic alignment in text-supervised models.

Abstract

Recently, semantic segmentation models trained with image-level text supervision have shown promising results in challenging open-world scenarios. However, these models still face difficulties in learning fine-grained semantic alignment at the pixel level and predicting accurate object masks. To address this issue, we propose MixReorg, a novel and straightforward pre-training paradigm for semantic segmentation that enhances a model's ability to reorganize patches mixed across images, exploring both local visual relevance and global semantic coherence. Our approach involves generating fine-grained patch-text pairs data by mixing image patches while preserving the correspondence between patches and text. The model is then trained to minimize the segmentation loss of the mixed images and the two contrastive losses of the original and restored features. With MixReorg as a mask learner, conventional text-supervised semantic segmentation models can achieve highly generalizable pixel-semantic alignment ability, which is crucial for open-world segmentation. After training with large-scale image-text data, MixReorg models can be applied directly to segment visual objects of arbitrary categories, without the need for further fine-tuning. Our proposed framework demonstrates strong performance on popular zero-shot semantic segmentation benchmarks, outperforming GroupViT by significant margins of 5.0%, 6.2%, 2.5%, and 3.4% mIoU on PASCAL VOC2012, PASCAL Context, MS COCO, and ADE20K, respectively.