Advances in Semantic Segmentation

The field of semantic segmentation is rapidly advancing with a focus on addressing limitations in few-shot segmentation, weakly supervised semantic segmentation, and out-of-distribution segmentation. Researchers are exploring novel approaches to improve the accuracy and generalizability of segmentation models, including the use of category-specific high-level representations, directional masking strategies, and objectness-aware refinement frameworks. Noteworthy papers in this area include I$^2$R, which proposes a novel few-shot segmentation method that achieves state-of-the-art performance on benchmark datasets. Another notable paper is Objectomaly, which introduces an objectness-aware refinement framework for out-of-distribution segmentation and achieves state-of-the-art performance on key benchmarks. Additionally, the paper Know Your Attention Maps presents an end-to-end method for weakly supervised semantic segmentation that utilizes attention maps learned by a Vision Transformer, resulting in accurate pseudo-masks and comparable performance to fully-supervised models.

Sources

I$^2$R: Inter and Intra-image Refinement in Few Shot Segmentation

Know Your Attention Maps: Class-specific Token Masking for Weakly Supervised Semantic Segmentation

Adaptive Part Learning for Fine-Grained Generalized Category Discovery: A Plug-and-Play Enhancement

Objectomaly: Objectness-Aware Refinement for OoD Segmentation with Structural Consistency and Boundary Precision

Diffusion-Guided Knowledge Distillation for Weakly-Supervised Low-Light Semantic Segmentation

Rethinking Query-based Transformer for Continual Image Segmentation

Built with on top of