开源代码地址： https://github.com/Weizhi-Zhong/IP_LAP
conda create -n iplap Python 3.7.13
conda activate iplap
pip install torch torchvision torchaudio // 如果不行，就在官网查看安装命令https://pytorch.org/
pip install face-alignment==1.3.4
pip install -r requirements.txt
conda install ffmpeg
Download the pre-trained models from OneDrive or jianguoyun, and place them to the folder test/checkpoints . Then run the inference command.
预训练模型备份：
local_oss/IP_LAP/renderer_checkpoint.pth
local_oss/IP_LAP/landmarkgenerator_checkpoint.pth
The evaluation code is similar to this repo.
CUDA_VISIBLE_DEVICES=0 python inference_single.py # 测试
python inference_single.py --input './test/template_video/129.mp4' --audio './upload/t1.m4a' --landmark_gen_checkpoint_path './test/checkpoints/landmarkgenerator_checkpoint.pth'