Unleashing the Potential of Segment Anything Model 2 for Efficient ORSI Salient Object Detection

Clone Repository

git clone https://github.com/wellMachine/SAM2-RS.git
cd SAM2-RS/

Requirements

This project uses the SAM2 code in ./sam2/ and does not require installing SAM2 as a separate package. If you already have a working environment for SAM2, you can reuse it. Otherwise, you may create a new conda environment:

conda create -n sam2-rs python=3.10
conda activate sam2-rs
pip install -r requirements.txt

Project Structure

• SAM2-RS.py: model implementation (contains class SAM2_RS)
• SAM2_RS.py: import bridge so you can do from SAM2_RS import SAM2_RS
• train.py: training script
• test.py: inference script (saves predicted masks)
• eval.py: evaluation script (metrics)
• train.sh, test.sh, eval.sh: runnable examples with placeholder paths
• dataset.py, metrics.py: dataset/metric utilities
• sam2/: SAM2-related code used by this project

Requirements

This project uses the SAM2 code in ./sam2/ and does not require installing SAM2 as a separate package. If you already have a working environment for SAM2, you can reuse it. Otherwise, you may create a new conda environment:

conda create -n sam2-rs python=3.10
conda activate sam2-rs
pip install -r requirements.txt

Training

Download Weights

Download the pretrained weights file sam2_hiera. The download link is wrapped below:

• sam2_hiera_tiny.pt
• sam2_hiera_small.pt
• sam2_hiera_base_plus.pt
• sam2_hiera_large.pt

Run Training

Use the train.sh script to start training. Make sure the weights file is placed in the correct path (e.g., sam2_hiera_large.pt in the current directory).

Example train.sh content:

#!/bin/bash
# Training script example
python train.py \
--hiera_path "/path/to/sam2_hiera_large.pt" \
--train_image_path "/path/to/train/images/" \
--train_mask_path "/path/to/train/masks/" \
--save_path "/path/to/output/checkpoints/" \
--epoch  \
--lr  \
--batch_size

Test

Use the test.sh script to run testing.

Run Testing

Example test.sh content:

#!/bin/bash
# Testing script example
CUDA_VISIBLE_DEVICES="" \
python test.py \
--checkpoint "/path/to/checkpoints/SAM2_RS-best.pth" \
--test_image_path "/path/to/test/images/" \
--test_gt_path "/path/to/test/masks/" \
--save_path "/path/to/output/predictions/" 

Eval

Use the eval.sh script to run testing.

Run Eval

Example eval.sh content:

#!/bin/bash
# Eval script example
python eval.py \
--pred_path "/path/to/prediction/results/" \
--gt_path "/path/to/ground_truth/masks/" 

📥 预测图下载

预测结果已上传至百度网盘（提取码：t4dv），点击下面链接下载：
链接: https://pan.baidu.com/s/1lYovmYvIstcomPTlPVvbjQ?pwd=t4dv 提取码: t4dv