Project page: http://www.wisdom.weizmann.ac.il/~vision/ingan/ (See our results and visual comparison to other methods)

If you find our work useful in your research or publication, please cite our work:

@InProceedings{InGAN,
  author = {Assaf Shocher and Shai Bagon and Phillip Isola and Michal Irani},
  title = {InGAN: Capturing and Retargeting the "DNA" of a Natural Image},
  booktitle = {The IEEE International Conference on Computer Vision (ICCV)},
  year = {2019}
}

First you have to download the example checkpoint file, and put it in InGAN/examples/fruit/. Will defaulty run on the fruits image, using an existing checkpoint.

python test.py

By default, when testing you get a collage of various sizes and a smooth video of the transforms. You can also choose to test specific sizes, non-rectangular transforms and more.

See configs.py, for all the options. You can either edit this file or modify configuration from command-line. Examples:

python test.py --input_image_path /path/to/some/image.png  # choose input image
python test.py --test_non_rect  # also output non rectangular transformation results
python test.py --test_vid_scale 2.0, 0.5, 2.5, 0.2  # boundary scales for output video: [max_v, min_v, max_h, min_h]

Please see configs.py for many more options

Will defaulty run on the fruits image.

python train.py

See configs.py for all the options. You can either edit this file or modify configuration from command-line. Examples:

python train.py --input_image_path /path/to/some/image.png  # choose input image
python train.py --G_num_resblocks 3  # change number of residual block in the generator

Please see configs.py for many more options

In you results folder, monitor files will be periodically created, example:

Please see the file supp_video.py

Please see the file train_supp_mat.py