*Pingping Zhang✉️1,Tianyu Yan1,Yang Liu1, Huchuan Lu1
Dalian University of Technology, IIAU-Lab1
As an important pillar of underwater intelligence, Marine Animal Segmentation (MAS) involves segmenting animals within marine environments. Previous methods don't excel in extracting long-range contextual features and overlook the connectivity between pixels. Recently, Segment Anything Model (SAM) offers a universal framework for general segmentation tasks. Unfortunately, trained with natural images, SAM does not obtain the prior knowledge from marine images. In addition, the single-position prompt of SAM is very insufficient for prior guidance. To address these issues, we propose a novel learning framework, named Dual-SAM for high-performance MAS. To this end, we first introduce a dual structure with SAM's paradigm to enhance feature learning of marine images. Then, we propose a Multi-level Coupled Prompt (MCP) strategy to instruct comprehensive underwater prior information, and enhance the multi-level features of SAM's encoder with adapters. Subsequently, we design a Dilated Fusion Attention Module (DFAM) to progressively integrate multi-level features from SAM's encoder. With dual decoders, it generates pseudo-labels and achieves mutual supervision for harmonious feature representations. Finally, instead of directly predicting the masks of marine animals, we propose a Criss-Cross Connectivity Prediction ( C3P) paradigm to capture the inter-connectivity between pixels. It shows significant improvements over previous techniques. Extensive experiments show that our proposed method achieve state-of-the-art performances on five widely-used MAS datasets.
- [Dual-SAM] is a novel learning framework for high performance Marine Animal Segmentation (MAS). The framework inherits the ability of SAM and adaptively incorporate prior knowledge of underwater scenarios.
- Motivation of Our proposed Mehtod
- Multi-level Coupled Prompt
- Criss-Cross Connectivity Prediction
- Dilated Fusion Attention Module
We rely on five public datasets and five evaluation metrics to thoroughly validate our model’s performance.
step1:Clone the Dual_SAM repository:
To get started, first clone the Dual_SAM repository and navigate to the project directory:
git clone https://github.com/Drchip61/Dual_SAM.git
cd Dual_SAM
step2:Environment Setup:
Dual_SAM recommends setting up a conda environment and installing dependencies via pip. Use the following commands to set up your environment:
conda create -n Dual_SAM
conda activate Dual_SAM
pip install -r requirements.txt
Please put the pretrained SAM model in the Dual-SAM file.
Training
# Change the hyper parameter in the train_s.py
python train_s.py
Testing
# Change the hyper parameter in the test_y.py
python test_y.py
# First threshold the prediction mask
python bimap.py
# Then evaluate the perdiction mask
python test_score.py
@inproceedings{
anonymous2024fantastic,
title={Fantastic Animals and Where to Find Them: Segment Any Marine Animal with Dual {SAM}},
author={Pingping Zhang,Tianyu Yan, Yang Liu,Huchuan Lu},
booktitle={Conference on Computer Vision and Pattern Recognition 2024},
year={2024}
}