S

Sam Hq Vit Huge

Developed by syscv-community
SAM-HQ is an enhanced version of the Segment Anything Model (SAM), capable of generating higher-quality object masks, especially suitable for handling objects with complex structures.
Downloads 516
Release Time : 5/5/2025

Model Overview

By introducing high-quality output tokens and global-local feature fusion techniques, SAM-HQ significantly improves the quality of segmentation masks while retaining the original SAM's promptable design, efficiency, and zero-shot generalization capabilities.

Model Features

High-Quality Output Tokens
Specially designed learnable tokens injected into the mask decoder to predict more accurate segmentation masks.
Global-Local Feature Fusion
Fuses mask decoder features with early and final ViT features, combining high-level semantics and low-level boundary information to improve mask details.
Efficient Improvements
Adds less than 0.5% parameters and requires only 4 hours of training on 8 GPUs to significantly enhance segmentation quality.
Zero-Shot Generalization
Maintains the original SAM's zero-shot generalization capability, allowing direct application to unseen data.

Model Capabilities

High-quality image segmentation
Prompt-based segmentation (points, boxes, etc.)
Automatic mask generation
Zero-shot transfer learning

Use Cases

Image Editing
Precise Object Extraction
Accurately segments objects from complex backgrounds, preserving details and thin structures.
Compared to the original SAM, it better preserves object boundary details.
Automated Annotation
High-Quality Data Annotation
Automatically generates precise object masks for training data annotation.
Reduces manual annotation workload and improves annotation quality.
Medical Image Analysis
Medical Structure Segmentation
Segments fine structures in medical images.
Suitable for medical applications requiring high-precision segmentation.
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
Š 2025AIbase