Video caption entailment improves concept based video similarity scoring
Video-Language Entailment for Video Similarity Estimation
Computer Vision and Pattern Recognition
Summary
Comparing how similar two videos are, especially based on specific ideas like actions or places, is hard for current AI models. The paper introduces a method called VICE that looks at each video paired with the other's caption to see how well they match for certain concepts. It then combines these checks from both directions to decide how similar the videos are and gives a short text explaining why. This approach helps the model reason about videos one at a time but still compare them deeply. Tests show this method works better than looking at both videos together.
What this means in practice
- •For video platform engineers: Improve video recommendation by measuring similarity for specific concepts like activities or locations.
- •For video metadata teams: Generate textual explanations of video similarity to assist in organizing and tagging large video collections.
Authors
Benedetta Liberatori, Lorenzo Vaquero, Elisa Ricci, Paolo Rota
Abstract
Comparing videos to assess their semantic similarity remains a challenging task for current Large Multimodal Models (LMMs), which are primarily trained with single-video objectives. While most prior works have focused on computing global similarity, recent studies have introduced the more challenging problem of concept-based similarity, which measures how similar two videos are with respect to specific concepts (e.g., activity, location). However, no dedicated approaches have yet leveraged LMMs to explicitly address this task. In this paper, we aim to fill this gap introducing Video Caption Entailment for video similarity estimation (VICE): a dual-branch design that leverages LMMs for cross-video reasoning. Each branch jointly processes one video and the caption of the other, measuring their alignment for a specific concept via video-language entailment. The outputs from both branches are fused to produce a final similarity score, along with a textual explanation of the model reasoning. This approach leverages the model's native single-video understanding while enabling fine-grained comparative reasoning between video and text. Experiments on ConViS-Bench and StepDiff show that VICE outperforms models that process both videos simultaneously, highlighting the promise of single-video reasoning for concept-based similarity estimation.