S

Slimsam Uniform 77

Developed by Zigeng
SlimSAM is an innovative SAM model compression method that efficiently reuses pre-trained SAM through a unified pruning-distillation framework, eliminating the need for extensive repeated training.
Downloads 18.82k
Release Time : 1/8/2024

Model Overview

SlimSAM is a lightweight version of the Segment Anything Model (SAM), achieving model compression through an innovative pruning-distillation framework, significantly reducing parameters and computational load while maintaining performance close to the original model.

Model Features

Efficient Compression
Achieves model compression with only 0.1% of training data, reducing parameters to 0.9% and computational load to 0.8%.
Alternate Slimming Strategy
Enhances knowledge transfer from the original SAM through progressive pruning and distillation steps.
Label-Free Pruning Criterion
Aligns pruning objectives with optimization directions, improving post-pruning distillation effects.

Model Capabilities

Image Segmentation
Object Recognition
Semantic Segmentation

Use Cases

Computer Vision
Object Segmentation
Accurately segments specific objects in images.
Maintains segmentation accuracy close to the original SAM with minimal training data.
Lightweight Deployment
Deploys image segmentation models on resource-constrained devices.
Significantly reduces model size and computational requirements.
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
© 2025AIbase