EventCoT: Event-centric Video Chain-of-thought for Reasoning Temporal Localization

2026-07-06

AI Summary

This work addresses the challenge of reasoning-based temporal localization, which requires jointly generating an answer and its corresponding time interval in videos. The paper proposes the first event-centric video chain-of-thought framework, which compresses visual input through event-level video tokenization, performs multi-step reasoning within each event to generate answers, and aligns answer placeholders with visual representations via embedding matching for precise temporal localization. Key innovations include an event-centric chain-of-thought mechanism, end-to-end joint modeling of reasoning and localization, and strong zero-shot transfer capability. Experiments demonstrate that the method achieves state-of-the-art performance on ActivityNet-RTL and exhibits robust zero-shot reasoning ability on the ReXTime benchmark.

Abstract

Reasoning temporal localization (RTL) requires a model to generate an answer that itself contains the time interval supporting it, so high-level reasoning and precise temporal grounding must be produced jointly in a single response. To tackle this challenging task, we propose the first event-centric video chain-of-thought framework, dubbed EventCoT. EventCoT first performs event-centric tokenization of the input video to convert it into compact event tokens, enabling efficient identification of question-relevant events. It then reasons within the identified events to generate the answer, grounding the time interval via embedding matching that aligns placeholder tokens with visual embeddings. EventCoT achieves state-of-the-art results on ActivityNet-RTL for reasoning temporal localization while using substantially fewer visual tokens than previous work. To verify its general performance, we further evaluate EventCoT on the grounded video question answering benchmark ReXTime, where it attains strong zero-shot results.

Problem

• Reasoning Temporal Localization
• Temporal Grounding
• Video Question Answering
• Event-centric Reasoning

Innovation

• event-centric tokenization
• video chain-of-thought
• temporal grounding
• embedding matching
• reasoning temporal localization