PhysGame: Uncovering Physical Commonsense Violations in Gameplay Videos

TL;DR

This paper introduces PhysGame, a benchmark for evaluating physical commonsense violations in gameplay videos, and proposes PhysVLM, a knowledge-enhanced video LLM, to improve understanding of physical glitches.

Contribution

The paper presents PhysGame as a new benchmark, curates datasets for instruction tuning and preference optimization, and develops PhysVLM to advance physical commonsense reasoning in video LLMs.

Findings

Current open-source video LLMs underperform proprietary models.

PhysVLM achieves state-of-the-art results on PhysGame and general benchmarks.

Datasets and methods significantly improve physical commonsense understanding.

Abstract

Recent advancements in video-based large language models (Video LLMs) have witnessed the emergence of diverse capabilities to reason and interpret dynamic visual content. Among them, gameplay videos stand out as a distinctive data source, often containing glitches that defy physics commonsense. This characteristic renders them an effective benchmark for assessing the under-explored capability of physical commonsense understanding in video LLMs. In this paper, we propose PhysGame as a pioneering benchmark to evaluate physical commonsense violations in gameplay videos. PhysGame comprises 880 videos associated with glitches spanning four fundamental domains (i.e., mechanics, kinematics, optics, and material properties) and across 12 distinct physical commonsense. Through extensively evaluating various state-ofthe-art video LLMs, our findings reveal that the performance of current open-source video LLMs significantly lags behind that of proprietary counterparts. To bridge this gap, we curate an instruction tuning dataset PhysInstruct with 140,057 question-answering pairs to facilitate physical commonsense learning. In addition, we also propose a preference optimization dataset PhysDPO with 34,358 training pairs, where the dis-preferred responses are generated conditioned on misleading titles (i.e., meta information hacking), fewer frames (i.e., temporal hacking) and lower spatial resolutions (i.e., spatial hacking). Based on the suite of datasets, we propose PhysVLM as a physical knowledge-enhanced video LLM. Extensive experiments on both physical-oriented benchmark PhysGame and general video understanding benchmarks demonstrate the state-ofthe-art performance of PhysVLM.