PREGEN: Uncovering Latent Thoughts in Composed Video Retrieval

TL;DR

PREGEN is an efficient, zero-shot framework for composed video retrieval that leverages frozen vision-language models and lightweight encoding to outperform prior methods significantly.

Contribution

It introduces a novel approach that avoids fine-tuning VLMs, using hidden states for semantic embedding, leading to state-of-the-art results in CoVR tasks.

Findings

Surpasses prior methods with +27.23 and +69.59 in Recall@1

Demonstrates robustness across different VLM backbones

Exhibits strong zero-shot generalization to complex modifications

Abstract

Composed Video Retrieval (CoVR) aims to retrieve a video based on a query video and a modifying text. Current CoVR methods fail to fully exploit modern Vision-Language Models (VLMs), either using outdated architectures or requiring computationally expensive fine-tuning and slow caption generation. We introduce PREGEN (PRE GENeration extraction), an efficient and powerful CoVR framework that overcomes these limitations. Our approach uniquely pairs a frozen, pre-trained VLM with a lightweight encoding model, eliminating the need for any VLM fine-tuning. We feed the query video and modifying text into the VLM and extract the hidden state of the final token from each layer. A simple encoder is then trained on these pooled representations, creating a semantically rich and compact embedding for retrieval. PREGEN significantly advances the state of the art, surpassing all prior methods on standard CoVR benchmarks with substantial gains in Recall@1 of +27.23 and +69.59. Our method demonstrates robustness across different VLM backbones and exhibits strong zero-shot generalization to more complex textual modifications, highlighting its effectiveness and semantic capabilities.