Accessing Vision Foundation Models via ImageNet-1K

TL;DR

This paper introduces Proteus, a simple method to distill large vision foundation models into smaller, accessible models trained on ImageNet-1K, enabling broader research without access to original training data.

Contribution

Proteus provides a novel, data-efficient distillation approach that removes dataset bias and achieves competitive performance with significantly less training data.

Findings

Proteus-L/14 matches DINOv2-L/14 performance across 19 benchmarks.

Proteus outperforms larger models like CLIP-L/14 and SynCLR-L/14 with fewer training images.

The method enables training foundation models at ImageNet-level costs.

Abstract

Vision foundation models are renowned for the generalization ability due to massive training data. Nevertheless, they demand tremendous training resources, and the training data is often inaccessible, e.g., CLIP, DINOv2, posing great challenges to developing derivatives that could facilitate the research. In this work, we offer a very simple and general solution, named \textit{Proteus}, to distill foundation models into smaller equivalents on ImageNet-1K without access to the original training data. Specifically, we remove the designs from conventional knowledge distillation settings that result in dataset bias and present three levels of training objectives, i.e., token, patch, and feature, to maximize the efficacy of knowledge transfer. In this manner, Proteus is trained at ImageNet-level costs with surprising ability, facilitating the accessibility of training foundation models for the broader research community. When leveraging DINOv2-g/14 as the teacher, Proteus-L/14 matches the performance of the Oracle method DINOv2-L/14 (142M training data) across 19 benchmarks and outperforms other vision foundation models including CLIP-L/14 (400M), OpenCLIP-L/14 (400M/2B) and SynCLR-L/14 (600M) with a significantly smaller training set of 1.2M images.