anliyuan / Ultralight-Digital-Human

一个超轻量级、可以在移动端实时运行的数字人模型
934 stars 149 forks source link

在执行inference.py时报错 #66

Open drq123456 opened 5 days ago

drq123456 commented 5 days ago

(dh) (base) root@autodl-container-2d0241b899-289114e4:~/Ultralight-Digital-Human-master# python inference.py --asr wenet --dataset /root/metahumanvideo --audio_feat /root/metahumanvideo/aud_wenet.npy --save_path 1.mp4 --checkpoint /root/Ultralight-Digital-Human-master/syncnet_ckpt/39.pth OpenCV: FFMPEG: tag 0x47504a4d/'MJPG' is not supported with codec id 7 and format 'mp4 / MP4 (MPEG-4 Part 14)' OpenCV: FFMPEG: fallback to use tag 0x7634706d/'mp4v' Traceback (most recent call last): File "/root/Ultralight-Digital-Human-master/inference.py", line 64, in net.load_state_dict(torch.load(checkpoint)) File "/root/miniconda3/envs/dh/lib/python3.10/site-packages/torch/nn/modules/module.py", line 2189, in load_state_dict raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format( RuntimeError: Error(s) in loading state_dict for Model: 请问大家遇到过这种报错吗,我在执行inference.py文件推理时遇到的错误,求助

anliyuan commented 2 days ago

inference时候为什么要加载syncnet的模型啊