I feel like I’m missing something. Almost everything comes back with question mark. But then i search for ‘se’ by accident and the suggestion is ‘dachshundt-on-record-player-turntable’.
There are only 61 GIFs in the entire dataset [0] What this project is most clearly missing is an attempt at web crawling for GIFs. This is how Giphy works, and frankly I think it would be trivial to scrape a few thousand GIFs in an afternoon.
The data set is small. The search system is exclusively based on similarity search using word vectors. One result of this is that an image that should be returned for a search for 'parrot' is only returned for a search for 'bird'. A better search system would combine both word vectors and keyword matching so that if a description contains parrot it matches parrot but bird would also match parrot.
Might want to considering outsourcing the actual search functionality to an external engine for better accuracy. Here are some good choices:
https://github.com/topics/search-engine
13 comments
[ 4.3 ms ] story [ 38.5 ms ] threadI searched for "salty" then "salt" "wave". The suggestions were "question_mark", "faucet", and "ocean sunset" respectively.
[0] https://github.com/bnlcas/OpenGif/blob/master/Data/Gif_Data/...
https://www.google.com/search?q=meme&tbm=isch&tbs=itp:animat...
So far, this is nowhere near the quality of a google image search, or even giphy.