https://github.com/mmaaz60/object_detection_in_aerial_images

The repository contains the code for Object Detection in Aerial Images (iSAID dataset) using Faster RCNN and scale-aware data augmentation (SA-AutoAug).
https://github.com/mmaaz60/object_detection_in_aerial_images

Last synced: 18 days ago
JSON representation

The repository contains the code for Object Detection in Aerial Images (iSAID dataset) using Faster RCNN and scale-aware data augmentation (SA-AutoAug).

• Host: GitHub
• URL: https://github.com/mmaaz60/object_detection_in_aerial_images
• Owner: mmaaz60
• Created: 2022-04-10T13:30:07.000Z (over 2 years ago)
• Default Branch: main
• Last Pushed: 2022-04-10T13:55:38.000Z (over 2 years ago)
• Last Synced: 2024-10-10T18:11:04.815Z (about 1 month ago)
• Language: Python
• Size: 780 KB
• Stars: 4
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
## Installation

The code is tested with PyTorch 1.8.0 and CUDA 11.1.

1. Install PyTorch and torchvision

```shell

pip install torch==1.8.0+cu111 torchvision==0.9.0+cu111 -f https://download.pytorch.org/whl/torch_stable.html

```

2. Install Detectron2

```shell

python -m pip install -e .

```

## Training

Use script `train_faster_rcnn.py` to train the models. The script expect the following parameters,

* -data_dir -> iSAID dataset path

* -config -> Detectron2 config file listing all model and training related configurations

* -output_dir -> Output directory to save checkpoints and logs

* --resume -> Flag to resume the training from the available latest checkpoints

* --eval_only -> Flag used to perform only the evaluation

* --eval_checkpoints -> Path to the checkpoints to use for the evaluation

The configs for training using SA-AutoAug are available at [here](configs/SA_AutoAug).

## Evaluate pretrained models

Run the following command to evaluate the provided pretrained models,

```shell

python train_faster_rcnn.py -data_dir  -output_dir  --eval_only --eval_checkpoints 

```

## Visualization

The visualizations can be generated using the script `visualize_detections.py`.

Should you have any questions, please contact at [email protected] or [email protected]