PerceptionComp: A Video Benchmark for Complex Perception-Centric Reasoning

TL;DR

PerceptionComp is a new manually annotated video benchmark designed to evaluate complex perception-centric reasoning involving multiple visual and logical subtasks across diverse domains.

Contribution

It introduces a challenging benchmark with 1,114 questions on 279 videos, highlighting the difficulty of perception-centric long-horizon reasoning for both humans and AI models.

Findings

Humans take longer and perform worse when rewatching is disallowed.

State-of-the-art models achieve less than 46% accuracy, indicating a significant gap.

Perception-centric reasoning remains a major bottleneck for current AI systems.

Abstract

We introduce PerceptionComp, a manually annotated benchmark for complex, long-horizon, perception-centric video reasoning. PerceptionComp is designed so that no single moment is sufficient: answering each question requires multiple temporally separated pieces of visual evidence and compositional constraints under conjunctive and sequential logic, spanning perceptual subtasks such as objects, attributes, relations, locations, actions, and events, and requiring skills including semantic recognition, visual correspondence, temporal reasoning, and spatial reasoning. The benchmark contains 1,114 highly complex questions on 279 videos from diverse domains including city walk tours, indoor villa tours, video games, and extreme outdoor sports, with 100% manual annotation. Human studies show that PerceptionComp requires substantial test-time thinking and repeated perception steps: participants take much longer than on prior benchmarks, and accuracy drops to near chance (18.97%) when rewatching is disallowed. State-of-the-art MLLMs also perform substantially worse on PerceptionComp than on existing benchmarks: the best model in our evaluation, Gemini-3-Flash, reaches only 45.96% accuracy in the five-choice setting, while open-source models remain below 40%. These results suggest that perception-centric long-horizon video reasoning remains a major bottleneck, and we hope PerceptionComp will help drive progress in perceptual reasoning.