R

Robustsam Vit Large

Developed by jadechoghari
RobustSAM is a model for robust segmentation of arbitrary objects in degraded images, improved upon SAM with enhanced segmentation performance on low-quality images.
Downloads 86
Release Time : 8/16/2024

Model Overview

RobustSAM is an improved version of the Segment Anything Model (SAM), focusing on enhancing segmentation performance on quality-degraded images (e.g., blur, haze, low-light conditions) while retaining SAM's original zero-shot generalization capability and prompt system.

Model Features

Robustness on Degraded Images
Specially optimized to handle image segmentation tasks under degraded conditions like blur, haze, and low-light
Zero-shot Generalization
Retains SAM's zero-shot learning capability to handle new categories without specific training
Efficient Training
Requires only minimal parameter increases and can be optimized within 30 hours on 8 GPUs
Comprehensive Dataset Support
Provides the Robust-Seg dataset containing 688K degraded image-mask pairs

Model Capabilities

Image Segmentation
Zero-shot Learning
Prompt-based Segmentation
Automatic Mask Generation
Degraded Image Processing

Use Cases

Computer Vision
Degraded Image Segmentation
Image segmentation under conditions like blur, haze, and low-light
Significant performance improvement compared to original SAM
Single Image Dehazing/Deblurring
Acts as a preprocessing step to enhance downstream dehazing/deblurring tasks
Effectively improves downstream task performance
Autonomous Driving
Object Detection in Adverse Weather
Road object segmentation in adverse weather conditions like rain and fog
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
© 2025AIbase