HyperLips icon indicating copy to clipboard operation
HyperLips copied to clipboard

Pytorch official implementation for our paper "HyperLips: Hyper Control Lips with High Resolution Decoder for Talking Face Generation".

Results 13 HyperLips issues
Sort by recently updated
recently updated
newest added

My outputs are all shakey and the lips do not track properly. Even the provided example saved in the git is glitchy, what's going on? https://github.com/semchan/HyperLips/raw/main/results/result_voice.mp4 For anyone wanting to...

清晰度不错,嘴部纹理特征都保留了,但是嘴唇抖动厉害,不知道是不是syncnet训练的不够好

whether this model support the resolution of 512*512? if yes, how can i set the parameter of '--checkpoint_path_BASE' and '--checkpoint_path_HR'? ![pth](https://github.com/semchan/HyperLips/assets/78424820/6a327ae2-90a1-4e6c-85c6-a85c86ac9993) ![error](https://github.com/semchan/HyperLips/assets/78424820/f67823be-ca0f-4eae-904c-75a1b6a85fe9)

请问大佬,data_process_hyper_hq_module中的 vfile_o = path.join(args.hyperlips_train_dataset, "video_clips",(args.origin_data_root.split("/")[-1]),vidname+".mp4") 是为了覆盖前面生成的 video_clips里的视频吗?可是这里面的视频是为了分帧训练 lipsync expert用的,这些都是前几步的操作了,这里覆盖的用意何在,没想明白。

貌似这三个可以随便选,不知道网络结构要修改不? img_size=512, img_size=256, img_size=128,

Thanks for the great work, it is indeed interesting paper. Do you know the reason behind green dot on DINet you reproduced? while your model does not.