Bilallbayrakdar / ShipYOLO

Paper: ShipYOLO: An Enhanced Model for Ship Detection

DOI: https://doi.org/10.1155/2021/1060182

Xu Han, "ShipYOLO: An Enhanced Model for Ship Detection," Journal of Advanced Transportation, vol. 2021, Article ID 1060182, 11 pages, 2021. https://doi.org/10.1155/2021/1060182.

Bibtex

@article{
    author    = {Xu Han and Lining Zhao and Yue Ning and Jingfeng Hu}, 
    title     = {{ShipYOLO}: An Enhanced Model for Ship Detection}, 
    journal   = {Journal of Advanced Transportation},
    month     = {jun},  
    year      = {2021},
    pages     = {1--11}
    }

completed

• ShipYOLO
• Results for WSODD.

ongoing

• ShipYOLOv2
• Results for Seaships, SMD, et al.

Results and Models

Inference in NVIDIA GeForce RTX 3060 Laptop GPU.

Dataset

• WSODD

  DownLoad Link:https://github.com/sunjiaen/WSODD

  Train list and Test list for this paper:train.txt,test.txt

  The number of images in "datasets/WSODD" is only to give an example.

Model	Size	$mAP_{coco}$	$mAP_{coco}@50$	$mAP_{coco}@75$	FPS	Config	Datasets	Download
YOLOv4	512	0.251	0.582	0.173	34.50	mmdet	WSODD	------
ShipYOLO	512	0.296	0.615	0.237	44.96	mmdet	WSODD	------
ShipYOLOV2	512	0.296	0.617	0.241	44.20	mmdet	WSODD	------
-	512	0.423	0.780	0.391	44.20	mosaic+aug	WSODD	------

1. Installation related environment

pip install -r requirements.txt

2. Train

bash run_train.sh

python re_pth.py

3. demo

import torch
from models.shipyolo.shipyolo import ShipYOLOv2
from utils.general import non_max_suppression

'''
Train deploy=False
Test  deploy=True
'''
x = torch.rand(1, 3, 288, 512).cuda()
model = ShipYOLOv2(num_classes=14, deploy=True).cuda()
output = model(x)  # without nms
output = non_max_suppression(output[0], conf_thres=0.001, iou_thres=0.5)

4. structure

The old version of the code(ShipYOLO) is in "project/ShipYOLO".

We also built the code(ShipYOLO-mmdet) based on mmdetection which is in "project/ShipYOLO-mmdet".

Acknowledgements

• https://github.com/WongKinYiu/PyTorch_YOLOv4
• https://github.com/open-mmlab/mmdetection

Communication

• [e-mail]
• [e-mail]