AIbase
EN
Home
Tecoa4 Clip
T

Tecoa4 Clip

Developed by chs20
TeCoA is a vision-language model initialized from OpenAI CLIP, enhanced with supervised adversarial fine-tuning for improved robustness
Text-to-ImageOpen Source License:MIT#Adversarial Robustness#Zero-shot Classification#Image Understanding
Downloads 51
Release Time : 2/23/2024

Model Overview

This model is based on the CLIP ViT-L/14 architecture, adversarially fine-tuned on ImageNet using L∞ norm with radius 4/255 to enhance robustness against adversarial attacks. Primarily used for zero-shot image classification tasks.

Model Features

Adversarial Robustness
Significantly improves resistance to adversarial attacks through L∞ norm training with radius 4/255
Zero-shot Capability
Retains CLIP's zero-shot classification ability, applicable to novel categories without task-specific fine-tuning
Supervised Fine-tuning
Adversarial fine-tuning on ImageNet dataset balances accuracy and robustness

Model Capabilities

Zero-shot image classification
Adversarially robust image recognition
Cross-modal understanding (image-text)

Use Cases

Computer Vision
Safety-critical System Image Recognition
Reliable image classification in adversarial environments, suitable for autonomous driving, security systems
Maintains higher accuracy under adversarial attacks compared to standard CLIP
Open-domain Image Understanding
Leverages zero-shot capability to recognize unseen object categories
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
EnglishįŽ€äŊ“中文įšéĢ”中文ãĢãģんご
Š 2025AIbase