Scaling (Down) CLIP: A Comprehensive Analysis of Data, Architecture, and Training Strategies

TL;DR

This paper analyzes how data quality, model architecture, and training strategies impact the performance of scaled-down CLIP models, providing practical guidance for efficient training under limited compute resources.

Contribution

It offers a comprehensive analysis of CLIP scaling, highlighting the importance of high-quality data, suitable architecture choices, and effective training strategies for limited compute budgets.

Findings

High-quality data can outperform larger low-quality datasets.

Smaller ViT models are better for smaller datasets, larger models for bigger datasets.

CLIP+Data Augmentation achieves similar performance with half the data.

Abstract

This paper investigates the performance of the Contrastive Language-Image Pre-training (CLIP) when scaled down to limited computation budgets. We explore CLIP along three dimensions: data, architecture, and training strategies. With regards to data, we demonstrate the significance of high-quality training data and show that a smaller dataset of high-quality data can outperform a larger dataset with lower quality. We also examine how model performance varies with different dataset sizes, suggesting that smaller ViT models are better suited for smaller datasets, while larger models perform better on larger datasets with fixed compute. Additionally, we provide guidance on when to choose a CNN-based architecture or a ViT-based architecture for CLIP training. We compare four CLIP training strategies - SLIP, FLIP, CLIP, and CLIP+Data Augmentation - and show that the choice of training strategy depends on the available compute resource. Our analysis reveals that CLIP+Data Augmentation can achieve comparable performance to CLIP using only half of the training data. This work provides practical insights into how to effectively train and deploy CLIP models, making them more accessible and affordable for practical use in various applications.