Ditzley / DiffTED

GNU General Public License v3.0
9 stars 4 forks source link

Thank you so much for releasing code #1

Open johndpope opened 1 month ago

johndpope commented 1 month ago

are there any plans to release trained model? any idea on how long it will take to train?

i worked on recreating a paper recently from Microsoft - IMF- https://github.com/johndpope/imf it maybe of some interest - it throws away the keypoints and shifts focus to the decoder.

JP

UPDATE - question when you train the model - does it make most sense to split up the videos into scenes? eg. using pyscenedetect?

i draft some code here - https://gist.github.com/johndpope/7b0ec8c0329907dc13a9351a19bf20f2

iv2985 commented 2 weeks ago

@johndpope were you ever able to get anything running in this repo?

johndpope commented 2 weeks ago

I definitely chopped up videos successfully for training - was waiting to hear more before digging deeper. Also got distracted with mimo/ en3d - supposedly alobaba will open source code though not holding my breath