/AFN

Primary LanguagePython

Introduction

This repository contains the source code to replicate the results in our paper AFN: Attentional Feedback Network based 3D Terrain Super-Resolution at ACCV 2020, Kyoto.

System/Software Requirements

  1. PyTorch: 1.4
  2. Nvidia-1080Ti with 11GB of VRAM

Inference

  1. To get an inference with our pretrained model on your custom model, please add the test data pairs (aerial image and low-resolution DEM) in the datasets directory.
  2. Download the pretrained model from this path.
  3. Update the paths accordingly in options/test/test_options.json file.
  4. Use following command to test.
python test.py -opt options/test/test_options.py

Train

  1. To train the network on your own dataset, please add the training pairs into datasets directory.
  2. Update the paths accordingly in options/train/train_options.json file. You can play with the hyperparameters like number of steps: T, number of groups: N in the same file.
  3. For training, use follwing command:
python train.py -opt options/train/train_options.py

Citation

If you find this work useful in your research, please consider citing with: 
@inproceedings{kubade2020afn,
  title={AFN: Attentional Feedback Network based 3D Terrain Super-Resolution},
  author={Kubade, Ashish and Patel, Diptiben and Sharma, Avinash and Rajan, KS},
  booktitle={Proceedings of the Asian Conference on Computer Vision},
  year={2020}
}