VLM-KD: Knowledge Distillation from VLM for Long-Tail Visual Recognition

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zhang, Zaiwei, Meyer, Gregory P., Lu, Zhichao, Shrivastava, Ashish, Ravichandran, Avinash, Wolff, Eric M.
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912008480227328
author Zhang, Zaiwei
Meyer, Gregory P.
Lu, Zhichao
Shrivastava, Ashish
Ravichandran, Avinash
Wolff, Eric M.
author_facet Zhang, Zaiwei
Meyer, Gregory P.
Lu, Zhichao
Shrivastava, Ashish
Ravichandran, Avinash
Wolff, Eric M.
contents For visual recognition, knowledge distillation typically involves transferring knowledge from a large, well-trained teacher model to a smaller student model. In this paper, we introduce an effective method to distill knowledge from an off-the-shelf vision-language model (VLM), demonstrating that it provides novel supervision in addition to those from a conventional vision-only teacher model. Our key technical contribution is the development of a framework that generates novel text supervision and distills free-form text into a vision encoder. We showcase the effectiveness of our approach, termed VLM-KD, across various benchmark datasets, showing that it surpasses several state-of-the-art long-tail visual classifiers. To our knowledge, this work is the first to utilize knowledge distillation with text supervision generated by an off-the-shelf VLM and apply it to vanilla randomly initialized vision encoders.
format Preprint
id arxiv_https___arxiv_org_abs_2408_16930
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle VLM-KD: Knowledge Distillation from VLM for Long-Tail Visual Recognition
Zhang, Zaiwei
Meyer, Gregory P.
Lu, Zhichao
Shrivastava, Ashish
Ravichandran, Avinash
Wolff, Eric M.
Computer Vision and Pattern Recognition
For visual recognition, knowledge distillation typically involves transferring knowledge from a large, well-trained teacher model to a smaller student model. In this paper, we introduce an effective method to distill knowledge from an off-the-shelf vision-language model (VLM), demonstrating that it provides novel supervision in addition to those from a conventional vision-only teacher model. Our key technical contribution is the development of a framework that generates novel text supervision and distills free-form text into a vision encoder. We showcase the effectiveness of our approach, termed VLM-KD, across various benchmark datasets, showing that it surpasses several state-of-the-art long-tail visual classifiers. To our knowledge, this work is the first to utilize knowledge distillation with text supervision generated by an off-the-shelf VLM and apply it to vanilla randomly initialized vision encoders.
title VLM-KD: Knowledge Distillation from VLM for Long-Tail Visual Recognition
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2408.16930