Apply DP-GAN on Generative-AI Navigation Information Competition for UAV Reconnaissance in Natural Environments

We are TEAM_5101, participants in this competition. By combining images generated from the DP-GAN and @Tianming8585's PITI models, we achieved 6th place (FID: 88.8781) on the public dataset and 4th place (FID: 89.0964) on the private dataset. This repository contains the code for DP-GAN and the ensemble model.

Report Ensemble

Environment

Based on tensorflow:2.15.0-gpu, and install torch, pandas, opencv-python and wandb additionally.

Preprocessing

Please check UAV.py.

Training

python train.py --name UAV_instance_512_z128_400 --dataset_mode UAV_34_train --param_free_norm instance --z_dim 128 --num_epochs 400 --gpu_ids 0 --batch_size 8

Generate Images

python test.py --name UAV_instance_512_z128_400 --dataset_mode UAV_34_public --param_free_norm instance --z_dim 128 --ckpt_iter best --results_dir private --gpu_ids 0 --batch_size 8

Export the Output of D-net

python eval_pri.py --name UAV_instance_512_z128_400 --dataset_mode UAV_34_public --param_free_norm instance --z_dim 128 --gpu_ids 0 --ckpt_iter best --results_dir private

The command will output the average values of the fake, background, RI, and RO classes inferred by the D-net for each generated image from different models. Ensure the folder path for the generated images is correctly set before use.

Ensemble

Use ensemble.ipynb to set the ensemble rules and combine images generated by different models. Ensure that the outputs from the D-net are exported before use.

Others

More command examples can be found in UAV.

Reference

sj-li/DP_GAN

@inproceedings{schonfeld_sushko_iclr2021,
  title={Dual Pyramid Generative Adversarial Networks for Semantic Image Synthesis},
  author={Shijie Li, Ming-Ming Cheng, Juergen Gall},
  booktitle={British Machine Vision Conference},
  year={2022}
}

Tianming8585/PITI

@misc{
  title  = {Apply DP-GAN on Generative-AI Navigation Information Competition for UAV Reconnaissance in Natural Environments},
  author = {Wei-Chun Tsao},
  url    = {https://github.com/Tsao666/DP_GAN},
  year   = {2024}
}

Acknowledgement

This code is based on DP_GAN

Citation

If you use this work please cite

@misc{
  title  = {Apply DP-GAN on Generative-AI Navigation Information Competition for UAV Reconnaissance in Natural Environments},
  author = {Wei-Chun Tsao},
  url    = {https://github.com/Tsao666/DP_GAN},
  year   = {2024}
}