Prerequisites

DLAHSD: Dynamic label adopted in auxiliary head for SAR detection. An efficient SAR-ship detector based on center point. Our work improvements and ablation experimental results all conducted on the MMDetection toolkit. MMDetection works on Linux, Windows. It requires Python 3.7+, CUDA 9.2+ and PyTorch 1.5+.

Installation

Step 0. Install VS2019 for C++ compiler.

Step 1. Conda environment installmen and activation.

conda create -n new_env python=3.8
conda activate new_env

Step 2. install Pytorch and other useful python pakage

pip install torch==1.8.2+cu111 torchvision==0.9.2+cu111 torchaudio===0.8.2 -f https://download.pytorch.org/whl/lts/1.8/torch_lts.html
-i http://mirrors.aliyun.com/pypi/simple/ --trusted-host mirrors.aliyun.com

pip install cython matplotlib opencv-python timm -i [http://mirrors.aliyun.com/pypi/simple/](http://mirrors.aliyun.com/pypi/simple/) 
--trusted-host mirrors.aliyun.com

Step 3. Install MMCV.

pip install mmcv-full=={mmcv_version} -f https://download.openmmlab.com/mmcv/dist/{cu_version}/{torch_version}/index.html

Step 4. Install MMDetection.

git clone https://github.com/open-mmlab/mmdetection.git
cd mmdetection
pip install -v -e .

Dataset

SAR Ship Detection Dataset (SSDD): Official Release and Comprehensive Data Analysis https://drive.google.com/file/d/1glNJUGotrbEyk43twwB9556AdngJsynZ/view?usp=sharing https://pan.baidu.com/s/1Lpg28ZvMSgNXq00abHMZ5Q password: 2021

SSDD is organized in coco form and stored in DLAHSD\Official-SSDD-OPEN\BBox_SSDD\coco_style

Train

DLAHSD training shell

cd ../DLAHSD
python tools/train.py work_dirs/DLAHSD/DHALSD.py

Test

python tools/test.py work_dirs/DLAHSD/DHALSD.py work_dirs/DLAHSD/____.pth  --out='   '

Inference