ReasonSeg

ReasonSeg is a dataset used to evaluate text-guided object segmentation by assessing models' abilities to perform multi-turn reasoning and refine segmentation masks based on visual feedback.

Papers using ReasonSeg (6)

Efficient Reasoning via Thought Compression for Language Segmentation2026Discriminative Perception via Anchored Description for Reasoning Segmentation2026GenSeg-R1: RL-Driven Vision-Language Grounding for Fine-Grained Referring Segmentation2026RSAgent: Learning to Reason and Act for Text-Guided Segmentation via Multi-Turn Tool Invocations2025RSVP: Reasoning Segmentation via Visual Prompting and Multi-modal Chain-of-Thought2025Coprs: Learning Positional Prior From Chain-of-thought For Reasoning Segmentation2025