MetaCLIP-B16-FullCC2.5B Open-Source Model - Used to Reveal CLIP Training Data Screening and Boost Data Applications

Metaclip B16 Fullcc2.5b

Developed by facebook

MetaCLIP is an implementation of the CLIP framework applied to CommonCrawl data, aiming to reveal CLIP's training data filtering methods

Downloads 90.78k

Release Time : 10/9/2023

Model Overview

This model constructs a shared image-text embedding space, supporting tasks such as zero-shot image classification and text-based image retrieval

Data Transparency

First public disclosure of data preprocessing pipeline for CLIP-style models

Large-scale Training

Trained on 2.5 billion data points from CommonCrawl

Multimodal Capability

Simultaneously processes visual and textual information

Zero-shot image classification

Text-based image retrieval

Image-based text retrieval

Cross-modal embedding learning

Content Retrieval

Music Scene Recognition

Identifies music-related scene elements in images

Can distinguish scene labels like 'playing music' and 'doing sports'

Multimodal Applications

Image-Text Matching System

Builds association systems between images and descriptive texts

Property	Details
Model Type	MetaCLIP model, base-sized version, patch resolution 16
Training Data	2.5 billion data points of CommonCrawl (CC)

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base