search

sailordiary / LipNet-PyTorch

"LipNet: End-to-End Sentence-level Lipreading" in PyTorch
BSD 3-Clause "New" or "Revised" License
65 stars 20 forks source link

请教 end-to-end audiovisual speech recognition的.npy的文件生成问题 #1

Closed wanli4717 closed 5 years ago

wanli4717 commented 5 years ago

我是天津大学的一名研究生，，目前在复现end-to-end audiovisual speech recognition 论文实验结果，我看见了您在博主的git上留言了，并发现了其中的错误，我想找您请教一下那篇论文生成.npy的文件问题

sailordiary commented 5 years ago

Hi,

You can refer to https://github.com/mpc001/end-to-end-lipreading/issues/6#issuecomment-449582378 . The expected format is a numpy array of shape [29, 96, 96]. If you want to save the videos as npy files just do an np.save on the array of stacked frames (i.e. batch_feats).