hundredblocks / semantic-search

Semantic search for images and words using neural networks.
203 stars 72 forks source link

Faster similarity retrieval method? #5

Open zhangyanbo2007 opened 5 years ago

zhangyanbo2007 commented 5 years ago

I now have 140000 images to do similarity retrieval, annoy measured to about 10s, would like to ask if there is a faster retrieval method?

navneetkrc commented 5 years ago

You might be using a lot of features, ANNOY works better if the number of features is less than 100, Since the number of images is very high, you can try optimizing by reducing the vector dimensions. From summary of Spotify ANNoy features:- Works better if you don't have too many dimensions (like <100) but seems to perform surprisingly well even up to 1,000 dimensions.