SAM2-UNet: Segment Anything 2 Makes Strong Encoder for Natural and Medical Image Segmentation

TL;DR

SAM2-UNet leverages the powerful SAM2 encoder within a U-shaped framework, achieving superior performance across diverse image segmentation tasks without complex modifications.

Contribution

This work introduces SAM2-UNet, a versatile segmentation framework that effectively utilizes SAM2 as an encoder with parameter-efficient fine-tuning, outperforming existing methods.

Findings

Outperforms state-of-the-art methods on multiple segmentation tasks

Demonstrates the effectiveness of SAM2 as a strong encoder

Shows versatility across natural and medical image segmentation

Abstract

Image segmentation plays an important role in vision understanding. Recently, the emerging vision foundation models continuously achieved superior performance on various tasks. Following such success, in this paper, we prove that the Segment Anything Model 2 (SAM2) can be a strong encoder for U-shaped segmentation models. We propose a simple but effective framework, termed SAM2-UNet, for versatile image segmentation. Specifically, SAM2-UNet adopts the Hiera backbone of SAM2 as the encoder, while the decoder uses the classic U-shaped design. Additionally, adapters are inserted into the encoder to allow parameter-efficient fine-tuning. Preliminary experiments on various downstream tasks, such as camouflaged object detection, salient object detection, marine animal segmentation, mirror detection, and polyp segmentation, demonstrate that our SAM2-UNet can simply beat existing specialized state-of-the-art methods without bells and whistles. Project page: \url{https://github.com/WZH0120/SAM2-UNet}.