Satellite Remote Sensing Image Object Detection

This project is about detecting objects in satellite remote sensing images using RefineDet with PyTorch.

Framework

Author

• YPW
• Email: 1043773117@qq.com

Requirements

• Python 3
• PyTorch 0.4 or higher
• torchvision
• visdom
• numpy

Dataset

DOTA dataset is a large-scale dataset for object detection in aerial images, containing images from different sensors and platforms. Each image has a pixel size ranging from 800 × 800 to 20,000 × 20,000, and contains objects with various sizes, orientations, and shapes. The dataset has 18 common categories, such as boat, car, plane, etc. You can extend this demo to include more categories or use other models.

Usage

Training

To train the model, run the following command:

python train.py --dataset VOC --dataset_root *

You can also use other arguments to customize the training process, such as --resume to resume from a checkpoint, --visdom to use visdom for loss visualization, and --save_frequency to set the frequency of saving checkpoints.

Testing

To test the model, run the following command:

python test.py --dataset VOC --dataset_root * --trained_model pretrain/*.pth

You can also use other arguments to customize the testing process, such as --cuda to use CUDA for acceleration, and --voc_root to set the root directory of VOC dataset.

Results

The model achieves a mean average precision (mAP) of 74.2% on DOTA test set.

Demo