Generative-AI Navigation Information Competition for UAV Reconnaissance in Natural Environments I : Image Data Generation

Environment

Operating system: CentOS 7.8
Programmimg language: Python 3.8.19
Hardware: NVIDIA Tesla V100-PCIE-32GB

create folder

./
├── river
├── road

Intallation(conda environment)

An environment can be created with all the Python dependencies.

conda env create -f environment.yml

Data preparation

After the data preprocessing, place the training dataset img and label_img in train_B and train_A.

./river/AICUP2024-spring/dataset/cityscapes
├── train_A
│   ├── TRA_RI_1000000.png
│              .
|              .
|              .
│   └── TRA_RI_1002159.png
├── train_B
│   ├── TRA_RI_1000000.jpg
|              .
|              .
|              .
│   └── TRA_RI_1002159.jpg
├── test_A
|   ├── PRI_RI_1000000.png
|              . 
|              .
|              .
|   └── PRI_RI_1000359.png
└── test_B

./road/AICUP2024-spring/dataset/cityscapes
├── train_A
│   ├── TRA_RO_1002160.png
│              .
|              .
|              .
│   └── TRA_RO_1004319.png
├── train_B
│   ├── TRA_RO_1002160.jpg
|              .
|              .
|              .
│   └── TRA_RO_1004319.jpg
├── test_A
|   ├── PRI_RO_100360.png
|              . 
|              .
|              .
|   └── PRI_RO_1000719.png
└── test_B

Training

river

python train.py --name RIVER --label_nc 0 --no_instance

road

python train.py --name ROAD --label_nc 0 --no_instance

The model parameters will be saved in checkpoints folder.

latest_net_G.pth

latest_net_D.pth

Testing

river

python test.py --name RIVER --label_nc 0 --no_instance --how_many 360

road

python test.py --name ROAD --label_nc 0 --no_instance --how_many 360

The generated results will be saved in results folder. And it also can be downloaded from here