Image Translation with ASAPNets

Spatially-Adaptive Pixelwise Networks for Fast Image Translation, CVPR 2021

Webpage | Paper | Video

Installation

install requirements:

pip install -r requirements.txt

Code Structure

The code is heavily based on the official implementation of SPADE, and therefore has the saome structure:

• train.py, test.py: the entry point for training and testing.
• trainers/pix2pix_trainer.py: harnesses and reports the progress of training.
• models/pix2pix_model.py: creates the networks, and compute the losses.
• models/networks/: defines the architecture of all models.
• options/: creates option lists using argparse package. More individuals are dynamically added in other files as well. Please see the section below.
• data/: defines the class for loading images and label maps.

The ASAPNets generator is implementaed in:

• models/networks/generator: defines the architecture of the ASAPNets generator.

Dataset Preparation

facades

run:

cd data 
bash facadesHR_download_and_extract.sh

This will extract the facades full resolution images into datasets/facadesHR.

cityscapes

download the dataset into datasets/cityscapes and arrange in folders: train_images, train_labels, val_images, val_labels

Generating Images Using Pretrained Models

Pretraned models can be downloaded from here. Save the models under the checkpoints/ folder. Images can be generated using the command:

# Facades 512
bash test_facades512.sh

# Facades 1024
bash test_facades512.sh

# Cityscapes
bash test_cityscapes.sh

The outputs images will appear at the./results/ folder.

Training New Models

New models can be trained with the following commands. Prepare dataset in the ./datasets/ folder. Arrange in folders: train_images, train_labels, val_images, val_labels . For custom datasets, the easiest way is to use ./data/custom_dataset.py by specifying the option --dataset_mode custom, along with --label_dir [path_to_labels] --image_dir [path_to_images]. You also need to specify options such as --label_nc for the number of label classes in the dataset, --contain_dontcare_label to specify whether it has an unknown label, or --no_instance to denote the dataset doesn't have instance maps.

Run:

python train.py --name [experiment_name] --dataset_mode custom --label_dir [path_to_labels] -- image_dir [path_to_images] --label_nc [num_labels]

There are many additional options you can specify, please explore the ./options files. To specify the number of GPUs to utilize, use --gpu_ids.

Testing

Testing is similar to testing pretrained models.

python test.py --name [name_of_experiment] --dataset_mode [dataset_mode] --dataroot [path_to_dataset]

you can load the parameters used from training by specifying --load_from_opt_file.

Acknowledgments

This code is heavily based on the official implementation of SPADE. We thank the authors for sharing their code publicly!

License

Attribution-NonCommercial-ShareAlike 4.0 International (see file).

Citation

@inproceedings{RottShaham2020ASAP,
  title={Spatially-Adaptive Pixelwise Networks for Fast Image Translation},
  author={Rott Shaham, Tamar and Gharbi, Michael and Zhang, Richard and Shechtman, Eli and Michaeli, Tomer},
  booktitle={Computer Vision and Pattern Recognition (CVPR)},
  year={2021}
}