SurgLQA: Scalable Long-Horizon Surgical Video Question Answering

TL;DR

SurgLQA introduces a scalable framework for long-horizon surgical video question answering, enabling better modeling of extended procedural dynamics and causal dependencies in intraoperative settings.

Contribution

It proposes novel methods for long-range temporal reasoning and adaptive inference, and restructures a benchmark dataset for systematic evaluation.

Findings

Achieves consistent performance improvements in long-range reasoning tasks.

Demonstrates effectiveness on Colon-LQA and REAL-Colon-VQA datasets.

Abstract

Surgical Video Question Answering (VideoQA) provides a promising paradigm for dynamic intraoperative interpretation, enabling real-time decision support and context-aware retrieval in clinical environments. Nevertheless, existing approaches are predominantly restricted to images or short clips, limiting their ability to model long-range procedural dynamics and causal dependencies across extended surgical workflows. To address this challenge, we propose SurgLQA, a unified long-horizon VideoQA framework for scalable surgical reasoning. This framework incorporates Faithful Temporal Consolidation (FTC), which leverages intrinsic temporal cues to construct compact long-range representations while preserving fine-grained temporal fidelity. Further, we develop Temporally-Grounded Multi-Policy Scaling (TMS), an adaptive test-time inference paradigm that strategically adjusts policy-level reasoning capacity within temporally grounded contexts. To facilitate systematic evaluation, we restructured a long-duration colonoscopy VideoQA benchmark, Colon-LQA, and conducted extensive experiments on Colon-LQA and REAL-Colon-VQA. Experimental results demonstrate that our approach achieves consistent performance gains in long-range reasoning with temporally grounded inference. Code link: https://github.com/RascalGdd/SurgLQA.