E

Eva02 Small Patch14 224.mim In22k

Developed by timm
EVA02 feature/representation model, pretrained on ImageNet-22k via masked image modeling, suitable for image classification and feature extraction tasks.
Downloads 705
Release Time : 3/31/2023

Model Overview

The EVA-02 model is a Vision Transformer featuring mean pooling, SwiGLU, Rotary Position Embedding (ROPE), and is suitable for image classification and feature extraction tasks.

Model Features

Masked Image Modeling Pretraining
Pretrained on ImageNet-22k using EVA-CLIP as the MIM teacher, enhancing the model's representation capability.
Advanced Transformer Architecture
Incorporates mean pooling, SwiGLU, Rotary Position Embedding (ROPE), and other techniques to improve model performance.
Efficient Computation
With 21.6 million parameters and 6.1 GMACs, it is suitable for deployment in resource-constrained environments.

Model Capabilities

Image Feature Extraction
Image Classification

Use Cases

Computer Vision
Image Classification
Used for classifying images, supporting recognition of multiple categories.
Pretrained on ImageNet-22k, achieving high classification accuracy.
Feature Extraction
Extracts image feature representations for downstream tasks such as object detection and image retrieval.
Provides high-quality image feature representations.
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
Š 2025AIbase