Skip to content

Latest commit

 

History

History
71 lines (51 loc) · 3.31 KB

README.md

File metadata and controls

71 lines (51 loc) · 3.31 KB
Raw

THG 构建智能数字人

SadTalker

数字人生成可使用SadTalker（CVPR 2023）,详情介绍见 https://sadtalker.github.io

在使用前先下载SadTalker模型:

bash scripts/sadtalker_download_models.sh

Baidu (百度云盘) (Password: linl)

如果百度网盘下载，记住是放在checkpoints文件夹下，百度网盘下载的默认命名为sadtalker，实际应该重命名为checkpoints

Wav2Lip

数字人生成还可使用Wav2Lip（ACM 2020），详情介绍见 https://github.com/Rudrabha/Wav2Lip

在使用前先下载Wav2Lip模型：

Model Description Link to the model
Wav2Lip Highly accurate lip-sync Link
Wav2Lip + GAN Slightly inferior lip-sync, but better visual quality Link
Expert Discriminator Weights of the expert discriminator Link
Visual Quality Discriminator Weights of the visual disc trained in a GAN setup Link 
class Wav2Lip:
    def __init__(self, path = 'checkpoints/wav2lip.pth'):
        self.fps = 25
        self.resize_factor = 1
        self.mel_step_size = 16
        self.static = False
        self.img_size = 96
        self.face_det_batch_size = 2
        self.box = [-1, -1, -1, -1]
        self.pads = [0, 10, 0, 0]
        self.nosmooth = False
        self.device = 'cuda' if torch.cuda.is_available() else 'cpu'
        self.model = self.load_model(path)

    def load_model(self, checkpoint_path):
        model = wav2lip_mdoel()
        print("Load checkpoint from: {}".format(checkpoint_path))
        if self.device == 'cuda':
            checkpoint = torch.load(checkpoint_path)
        else:
            checkpoint = torch.load(checkpoint_path,
                                    map_location=lambda storage, loc: storage)
        s = checkpoint["state_dict"]
        new_s = {}
        for k, v in s.items():
            new_s[k.replace('module.', '')] = v
        model.load_state_dict(new_s)

        model = model.to(self.device)
        return model.eval()

ER-NeRF（Comming Soon）

ER-NeRF（ICCV2023）是使用最新的NeRF技术构建的数字人，拥有定制数字人的特性，只需要一个人的五分钟左右到视频即可重建出来，具体可参考 https://github.com/Fictionarry/ER-NeRF

后续会针对此更新