:fire: If MCNet is helpful in your photos/projects, please help to :star: it or recommend it to your friends. Thanks:fire:
[Paper [Project Page] [Poster Video]
Fa-Ting Hong, Dan Xu
The Hong Kong University of Science and Technology
https://github.com/harlanhong/ICCV2023-MCNET/assets/19970321/4e8af5f6-b042-4ced-af2c-93c95e1b7009
:triangular_flag_on_post: Updates
We now provide a clean version of MCNet, which does not require customized CUDA extensions.
Clone repo
git clone https://github.com/harlanhong/ICCV2023-MCNET.git
cd ICCV2023-MCNET
Install dependent packages
pip install -r requirements.txt
## Install the Face Alignment lib
cd face-alignment
pip install -r requirements.txt
python setup.py install
We take the paper version for an example. More models can be found here.
See config/vox-256.yaml
to get description of each parameter.
The pre-trained checkpoint of face depth network and our MCNet checkpoints can be found under following link: OneDrive.
Inference! To run a demo, download checkpoint and run the following command:
CUDA_VISIBLE_DEVICES=0 python demo.py --config config/vox-256.yaml --driving_video path/to/driving --source_image path/to/source --checkpoint path/to/checkpoint --relative --adapt_scale --kp_num 15 --generator Unet_Generator_keypoint_aware --result_video path/to/result --mbunit ExpendMemoryUnit --memsize 1
The result will be stored in path/to/result
. The driving videos and source images should be cropped before it can be used in our method. To obtain some semi-automatic crop suggestions you can use python crop-video.py --inp some_youtube_video.mp4
. It will generate commands for crops using ffmpeg.
1) VoxCeleb. Please follow the instruction from https://github.com/AliaksandrSiarohin/video-preprocessing.
To train a model on specific dataset run:
CUDA_VISIBLE_DEVICES=0 python -m torch.distributed.launch --master_addr="0.0.0.0" --master_port=12347 run.py --config config/vox-256.yaml --name MCNet --batchsize 8 --kp_num 15 --generator Unet_Generator_keypoint_aware --GFM GeneratorFullModel --memsize 1 --kp_distance 10 --feat_consistent 10 --generator_gan 0 --mbunit ExpendMemoryUnit
The code will create a folder in the log directory (each run will create a new name-specific directory).
Checkpoints will be saved to this folder.
To check the loss values during training see log.txt
.
By default the batch size is tunned to run on 8 GeForce RTX 3090 gpu (You can obtain the best performance after about 150 epochs). You can change the batch size in the train_params in .yaml
file.
Also, you can watch the training loss by running the following command:
tensorboard --logdir log/MCNet/log
When you kill your process for some reasons in the middle of training, a zombie process may occur, you can kill it using our provided tool:
python kill_port.py PORT
1) Resize all the videos to the same size e.g 256x256, the videos can be in '.gif', '.mp4' or folder with images. We recommend the later, for each video make a separate folder with all the frames in '.png' format. This format is loss-less, and it has better i/o performance.
2) Create a folder data/dataset_name
with 2 subfolders train
and test
, put training videos in the train
and testing in the test
.
3) Create a config config/dataset_name.yaml
, in dataset_params specify the root dir the root_dir: data/dataset_name
. Also adjust the number of epoch in train_params.
Our MCNet implementation is inspired by FOMM. We appreciate the authors of FOMM for making their codes available to public.
@inproceedings{hong23implicit,
title={Implicit Identity Representation Conditioned Memory Compensation Network for Talking Head video Generation},
author={Hong, Fa-Ting and Xu, Dan},
booktitle={ICCV},
year={2023}
}
@inproceedings{hong2022depth,
title={Depth-Aware Generative Adversarial Network for Talking Head Video Generation},
author={Hong, Fa-Ting and Zhang, Longhao and Shen, Li and Xu, Dan},
journal={IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
year={2022}
}
@inproceedings{hong2023depth,
title={DaGAN++: Depth-Aware Generative Adversarial Network for Talking Head Video Generation},
author={Hong, Fa-Ting and and Shen, Li and Xu, Dan},
journal={arXiv preprint arXiv:2305.06225},
year={2023}
}
If you have any question or collaboration need (research purpose or commercial purpose), please email fhongac@cse.ust.hk
.