Data or Language Supervision: What Makes CLIP Better than DINO?

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Liu, Yiming, Zhang, Yuhui, Ghosh, Dhruba, Schmidt, Ludwig, Yeung-Levy, Serena
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911208548859904
author Liu, Yiming
Zhang, Yuhui
Ghosh, Dhruba
Schmidt, Ludwig
Yeung-Levy, Serena
author_facet Liu, Yiming
Zhang, Yuhui
Ghosh, Dhruba
Schmidt, Ludwig
Yeung-Levy, Serena
contents CLIP outperforms self-supervised models like DINO as vision encoders for vision-language models (VLMs), but it remains unclear whether this advantage stems from CLIP's language supervision or its much larger training data. To disentangle these factors, we pre-train CLIP and DINO under controlled settings -- using the same architecture, dataset, and training configuration -- achieving similar ImageNet accuracy. Embedding analysis shows that CLIP captures high-level semantics (e.g., object categories, text), while DINO is more responsive to low-level features like colors and styles. When integrated into VLMs and evaluated on 20 VQA benchmarks, CLIP excels at text-intensive tasks, while DINO slightly outperforms on vision-centric ones. Variants of language supervision (e.g., sigmoid loss, pre-trained language encoders) yield limited gains. Our findings provide scientific insights into vision encoder design and its impact on VLM performance.
format Preprint
id arxiv_https___arxiv_org_abs_2510_11835
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Data or Language Supervision: What Makes CLIP Better than DINO?
Liu, Yiming
Zhang, Yuhui
Ghosh, Dhruba
Schmidt, Ludwig
Yeung-Levy, Serena
Computer Vision and Pattern Recognition
Artificial Intelligence
Computation and Language
Machine Learning
Multimedia
CLIP outperforms self-supervised models like DINO as vision encoders for vision-language models (VLMs), but it remains unclear whether this advantage stems from CLIP's language supervision or its much larger training data. To disentangle these factors, we pre-train CLIP and DINO under controlled settings -- using the same architecture, dataset, and training configuration -- achieving similar ImageNet accuracy. Embedding analysis shows that CLIP captures high-level semantics (e.g., object categories, text), while DINO is more responsive to low-level features like colors and styles. When integrated into VLMs and evaluated on 20 VQA benchmarks, CLIP excels at text-intensive tasks, while DINO slightly outperforms on vision-centric ones. Variants of language supervision (e.g., sigmoid loss, pre-trained language encoders) yield limited gains. Our findings provide scientific insights into vision encoder design and its impact on VLM performance.
title Data or Language Supervision: What Makes CLIP Better than DINO?
topic Computer Vision and Pattern Recognition
Artificial Intelligence
Computation and Language
Machine Learning
Multimedia
url https://arxiv.org/abs/2510.11835