v1: Learning to Point Visual Tokens for Multimodal Grounded Reasoning
This addresses the issue of visual grounding in multimodal reasoning for AI systems, representing an incremental advancement with a novel method for a known bottleneck.
The paper tackles the problem of multimodal models losing focus on relevant visual regions during long reasoning chains by introducing v1, a lightweight extension that enables active visual referencing through a point-and-copy approach, resulting in consistent performance improvements across various multimodal mathematical reasoning benchmarks.
When thinking with images, humans rarely rely on a single glance: they revisit visual information repeatedly during reasoning. However, existing models typically process images only once and thereafter generate reasoning entirely in text, lacking mechanisms to re-access or ground inference in visual representations. We empirically confirm this: as reasoning chains lengthen, models progressively lose focus on relevant regions. In response, we introduce v1, a lightweight extension that enables active visual referencing through a simple point-and-copy approach. This allows the model to identify relevant image patches and copy their embeddings back into the reasoning stream, ensuring that evolving hypotheses remain grounded in perceptual evidence. Crucially, our pointing strategy lets the MLLM directly select image patches using their semantic representations as keys, keeping perceptual evidence embedded in the same space as the model's reasoning. To train this capability, we construct v1g, a dataset of 300K multimodal reasoning traces with interleaved visual grounding annotations. Across various multimodal mathematical reasoning benchmarks, v1 consistently outperforms comparable baselines, establishing point-and-copy as a practical mechanism for grounded reasoning. The model checkpoint and dataset are available at github.com/jun297/v1.