Towards Storage-Efficient Visual Document Retrieval: An Empirical Study on Reducing Patch-Level Embeddings

TL;DR

This study explores methods to reduce memory usage in visual document retrieval by evaluating token pruning and merging, leading to a new model that significantly cuts memory while maintaining high retrieval performance.

Contribution

It introduces Light-ColPali/ColQwen2, a novel approach that effectively reduces patch-level embeddings with minimal performance loss, providing a practical baseline for efficient VDR.

Findings

Random token pruning outperforms sophisticated methods but is unsuitable for VDR.

Token merging can significantly reduce memory with minimal performance loss.

Light-ColPali/ColQwen2 maintains 98.2% of performance at 11.8% memory usage.

Abstract

Despite the strong performance of ColPali/ColQwen2 in Visualized Document Retrieval (VDR), it encodes each page into multiple patch-level embeddings and leads to excessive memory usage. This empirical study investigates methods to reduce patch embeddings per page at minimum performance degradation. We evaluate two token-reduction strategies: token pruning and token merging. Regarding token pruning, we surprisingly observe that a simple random strategy outperforms other sophisticated pruning methods, though still far from satisfactory. Further analysis reveals that pruning is inherently unsuitable for VDR as it requires removing certain page embeddings without query-specific information. Turning to token merging (more suitable for VDR), we search for the optimal combinations of merging strategy across three dimensions and develop Light-ColPali/ColQwen2. It maintains 98.2% of retrieval performance with only 11.8% of original memory usage, and preserves 94.6% effectiveness at 2.8% memory footprint. We expect our empirical findings and resulting Light-ColPali/ColQwen2 offer valuable insights and establish a competitive baseline for future research towards efficient VDR.