Not getting great semantic search results with this model
#5
by stephenmarsh - opened
Has anyone else had difficulty getting this model to yield decent results for semantic text searches of images? I put in the first 30,000 or so images from The Met's collection. So far my best result is "chair" giving me some chairs but also vases and tables.
A search for "horse" yields no horses, only seemingly random objects.
I spent many hours sanity-checking and debugging these issues and regenerating embeddings, but I can't get any better performance out of it.
Anyone experiencing similar or have any advice/guidance/wisdom on the situation?
Thanks in advance