Romain Beaumont
Romain Beaumont
we have an ongoing project at laion discord (https://discord.gg/eq3cAMZtCC) to try and make a good AudioClip and also to collect a larger text/audio dataset once these 2 bricks are available,...
or a small subsample of laion5B
I may give parquet with embeddings one more try to check if it's really that slow Alternative formats i can think of: * Tfrecord * orc ? * Some SQL...
Yeah you're right On Sat, Mar 19, 2022, 15:25 Allen Kamp ***@***.***> wrote: > Read me show is missing the "-" after .tar > > clip_inference --input_dataset "pipe:aws s3 cp...
(benefit of arrow compared to hdf5 is it's faster to write, otherwise same speed) also maybe finally make a package/guide out of the id to metadata methods
change the config file schema to support that
* [x] https://github.com/rom1504/clip-retrieval/issues/37
Ideas regarding filtering: Try doing soft filtering by adding/substracting terms/average of terms to the query. Relevant terms could even be proposed to the user by having a term index. (Similar...
> do a dot product with a list of interesting clip embeddings, display these keywords as common attribute of items, propose to the user to add that to his query...
> do a dot product with a list of interesting clip embeddings, display these keywords as common attribute of items, propose to the user to add that to his query...