Iterative Visual Thinking: Teaching Vision-Language Models Spatial Self-Correction through Visual Feedback
For researchers and practitioners using VLMs for spatial reasoning tasks, this work addresses the critical gap between grounding and self-correction, enabling iterative refinement with modest data and compute.
The paper identifies that vision-language models (VLMs) lack self-correction ability in spatial grounding, with naive iteration causing a 31 percentage point drop in accuracy. The proposed Iterative Visual Thinking (IVT) framework, using a two-phase training recipe (SFT + GRPO), improves Acc@0.5 from 79.6% to 82.0% on a mixed benchmark, using only 2,400 training samples on a single GPU.
Vision-language models (VLMs) achieve strong singleshot spatial grounding, yet lack any mechanism to observe and correct their own predictions. We find that naively prompting a VLM to iterate over rendered visualizations of its predictions causes catastrophic failure: Acc@0.5 on referring expression comprehension collapses from 79.6% to 48.7% (a 31 percentage point drop), revealing a fundamental gap between grounding capability and self-correction ability. We propose Iterative Visual Thinking (IVT), a closed-loop framework in which the model predicts a bounding box, observes the prediction rendered on the image, and iteratively refines through visual feedback. A two-phase training recipe closes the self-correction gap: first, we exploit the base model's own predictions as realistic errors and prompt a teacher VLM to generate corrective reasoning traces, yielding supervised data without human annotation; second, we apply Group Relative Policy Optimization (GRPO) with a simple IoU reward to stabilize multi-step refinement. On a mixed benchmark spanning RefCOCOg, Ref-Adv, and Ref-L4 (505 test samples), SFT warm-up with IVT surpasses the single-shot base model on every metric: Acc@0.5 rises to 82.0% (+2.4pp), Acc@0.7 to 74.1% (+3.2pp), and Acc@0.9 to 48.3% (+2.8pp). GRPO further reduces per-step IoU degradation by 5x, stabilizing the refinement trajectory. All training uses only 2,400 samples on a single GPU, demonstrating that spatial self-correction is a learnable capability that can be instilled at modest scale.