Leaner and Faster: Two-Stage Model Compression for Lightweight Text-Image Retrieval

TL;DR

This paper introduces a two-stage model compression framework that significantly reduces size and increases speed of pre-trained text-image retrieval models, making them more practical for mobile devices without sacrificing accuracy.

Contribution

The proposed two-stage compression method effectively reduces model size and improves processing speed while maintaining or enhancing retrieval performance.

Findings

Model size reduced to 39% of original

Processing speed increased by 1.6x for images and 2.9x for text

Achieves comparable or better accuracy on Flickr30K and MSCOCO

Abstract

Current text-image approaches (e.g., CLIP) typically adopt dual-encoder architecture using pre-trained vision-language representation. However, these models still pose non-trivial memory requirements and substantial incremental indexing time, which makes them less practical on mobile devices. In this paper, we present an effective two-stage framework to compress large pre-trained dual-encoder for lightweight text-image retrieval. The resulting model is smaller (39% of the original), faster (1.6x/2.9x for processing image/text respectively), yet performs on par with or better than the original full model on Flickr30K and MSCOCO benchmarks. We also open-source an accompanying realistic mobile image search application.