Open YodaEmbedding opened 6 months ago
Hi @YodaEmbedding, this has been marked as WIP for a while. Is it ready for merge?
From my limited testing, it's probably good.
However, I did want to test it more thoroughly when working on Ujwal's project to confirm.
SG, let me know!
Added
mode="video"
parameter toPreGeneratedMemmapDataset
andVimeo90kDataset
.To generate the pregenerated numpy datasets:
Example config (works only on latest CompressAI Trainer dev branch):