🛠️VidFace

(The official code)

📖 VidFace: A Full-Transformer Solver for Video FaceHallucination with Unaligned Tiny Snapshots

[Paper]

Abstract

In this paper, we investigate the task of hallucinating an authentic high-resolution (HR) human face from multiple low-resolution (LR) video snapshots. We propose a pure transformer-based model, dubbed VidFace, to fully exploit the full-range spatio-temporal information and facial structure cues among multiple thumbnails. Specifically, VidFace handles multiple snapshots all at once and harnesses the spatial and temporal information integrally to explore face alignments across all the frames, thus avoiding accumulating alignment errors. Moreover, we design a recurrent position embedding module to equip our transformer with facial priors, which not only effectively regularises the alignment mechanism but also supplants notorious pre-training. Finally, we curate a new large-scale video face hallucination dataset from the public Voxceleb2 benchmark, which challenges prior arts on tackling unaligned and tiny face snapshots. To the best of our knowledge, we are the first attempt to develop a unified transformer-based solver tailored for video-based face hallucination. Extensive experiments on public video face benchmarks show that the proposed method significantly outperforms the state of the arts.

BibTeX

@article{gan2021vidface,
  title={VidFace: A Full-Transformer Solver for Video FaceHallucination with Unaligned Tiny Snapshots},
  author={Gan, Yuan and Luo, Yawei and Yu, Xin and Zhang, Bang and Yang, Yi},
  journal={arXiv preprint arXiv:2105.14954},
  year={2021}
}

🔧 Dependencies and Installation

(This work is based on the framework of BasicSR)

Python >= 3.7
PyTorch >= 1.3
NVIDIA GPU + CUDA

Clone repo

git clone https://github.com/yuangan/VidFace.git

Install dependent packages

cd VidFace
pip install -r requirements.txt

Install VidFace

python setup.py develop

You may also want to specify the CUDA paths:

CUDA_HOME=/usr/local/cuda \
CUDNN_INCLUDE_DIR=/usr/local/cuda \
CUDNN_LIB_DIR=/usr/local/cuda \
python setup.py develop

VidFace has been tested on Linux and Windows with anaconda.

📦 Dataset Preparation

TUFS145K images can be downloaded from Google or [Baidu, access code: lxvd], then excute cat tufs145ka* > tufs145k.zip and extract it to VidFace-main fold.
TUFS145K landmarks can be downloaded from Google or [Baidu, access code: lxvd], download 'tufs145k_lmk_norm.pickle' and move it to './landmarks/'

Prepare your dataset

Please refer to DatasetPreparation.md for more details.

💻 Train and Test

Training and testing commands:

Training with One GPU:

    CUDA_VISIBLE_DEVICES=0 python basicsr/train.py -opt options/train/VidFace/vidface_h48_norm_l10.yml

Training with Multiple GPU:

    CUDA_VISIBLE_DEVICES=0,1,2,3 python -m torch.distributed.launch --nproc_per_node=4 --master_port=4327 basicsr/train.py -opt options/train/VidFace/vidface_h48_norm_l10.yml --launcher pytorch

If you want to get the result in our paper, plz use the tufs_train_val.txt in options/train/VidFace/vidface_final_h48_norm_l10.yml.

Testing with One GPU:

CUDA_VISIBLE_DEVICES=0 python basicsr/test.py -opt options/test/VidFace/test_tufs145k_final.yml

Testing with Multiple GPU:

CUDA_VISIBLE_DEVICES=0,1,2,3 python -m torch.distributed.launch --nproc_per_node=4 --master_port=4442 basicsr/test.py -opt options/test/VidFace/test_tufs145k_final.yml --launcher pytorch

If you want to get the result of IJBC, plz download 'IJBC' from above driver and extract 'IJBC_128_96_new.zip' to VidFace-main fold. Then test by relace options/test/VidFace/test_ijbc_final.yml with options/test/VidFace/test_tufs145k_final.yml.

🍇: Trained Model

If you don't want to train it by yourself, we provide a trained VidFace with 600000 iters now. you can download from above link in 'model' folder. Move 'net_g_600000.pth' to './experiments/' then you can get the result in our paper during testing.