FastSeg: Efficient Training-Free Open-Vocabulary Segmentation via Hierarchical Attention Refinement Method (2506.23323v1)

Abstract: Open-vocabulary semantic segmentation (OVSS) aims to segment objects from arbitrary text categories without requiring densely annotated datasets. Although contrastive learning based models enable zero-shot segmentation, they often lose fine spatial precision at pixel level, due to global representation bias. In contrast, diffusion-based models naturally encode fine-grained spatial features via attention mechanisms that capture both global context and local details. However, they often face challenges in balancing the number of iterations with the quality of the segmentation. In this work, we propose FastSeg, a novel and efficient training-free framework with only (1+1)-step of reverse process of a pretrained diffusion model (e.g., Stable Diffusion). Moreover, instead of running multiple times for different classes, FastSeg performs segmentation for all classes at once. To further enhance the segmentation quality, FastSeg introduces three key components: (i) a dual-prompt mechanism for discriminative, class-aware attention extraction, (ii) a Hierarchical Attention Refinement Method (HARD) that enhances fused cross-attention using scale-aligned selfattention maps, and (iii) a Test-Time Flipping (TTF) scheme designed to improve spatial consistency. Extensive experiments show that FastSeg achieves state-of-the-art training-free performance, obtaining 43.8% average mIoU across PASCAL VOC, PASCAL Context, and COCO Object benchmarks while maintaining superior inference efficiency. Our results demonstrate that FastSeg provides a strong foundation for extendability, bridging the gap between segmentation quality and inference efficiency.