search

MRzzm / DINet

The source code of "DINet: deformation inpainting network for realistic face visually dubbing on high resolution video."
895 stars 167 forks source link

hdtf dataset preprocess #97

Closed shanzhajuan closed 3 months ago

shanzhajuan commented 3 months ago

Hello, according to the hdtf clips processed by [https://github.com/MRzzm/HDTF], some clips have a length of 6min, may I ask how to deal with videos of different lengths? Cut a fixed length, or divide the sentence by hand?

Inferencer commented 3 months ago

it will split them when you start the training automatically, what it does it split them into 9 frames so if you have a video that is 25fps and 1 minute long it will create a folder for that video and inside that folder will now be 166 folders with 9 frames in each, and it will do it for every video