Repository cleanup of Blade6570/icface while I play around with it.
The general requirements are as follows:
- Python 3.7
- Pytorch 0.4.1.post2
- Visdom and dominate
- Natsort
The following command uses aws s3 to pull the pretrained model to src/checkpoints/gpubatch_resnet/
./data_get.sh getI've ported all this into a nice little conda environment for you to use
conda env create -f environment.yml
conda env create -f environment-dlib.yml
cd src# Activate dlib conda
conda activate icface-dlib
# Generate ./crop/1.png
python image_crop.py \
--image ./crop/test/trump.jpeg \
--id 1# Activate icface conda
conda activate icface
# Generate a sample video
python test.py \
--dataroot ./ \
--model pix2pix \
--which_model_netG resnet_6blocks \
--which_direction AtoB \
--dataset_mode aligned \
--norm batch \
--display_id 0 \
--batchSize 1 \
--loadSize 128 \
--fineSize 128 \
--no_flip \
--name gpubatch_resnet \
--how_many 1 \
--ndf 256 \
--ngf 128 \
--which_ref ./crop/<video-id>.png \
--gpu_ids 0 \
--csv_path ./crop/videos/<input-video>.csv \
--results_dir results_video
# Splice the audio
ffmpeg -i ./crop/out.mp4 -i ./crop/videos/<input-video>.mp4 -c copy -map 0:0 -map 1:1 -shortest ./crop/out_audio.mp4# Outside Container
docker run -it --rm algebr/openface:latest
# Outside Container (new terminal window)
docker cp src/crop/videos/<video-id>.mp4 <docker-container>:/home/openface-build
# Within container (/home/openface-build)
./build/bin/FeatureExtraction -f <video-id>.mp4
# Outside Container
docker cp <docker-container>:/home/openface-build/processed/<video-id>.csv src/crop/videos@article{tripathy+kannala+rahtu,
title={ICface: Interpretable and Controllable Face Reenactment Using GANs},
author={Tripathy, Soumya and Kannala, Juho and Rahtu, Esa},
journal={arXiv preprint arXiv:1904.01909},
year={2019}
}