ML Algorithm for picking the best picture

Python
1

Hi,

I want to create a tool for picking the best picture out of a big bunch of similar pictures.

Therefore I want to encode the pics to vectors, and store them in a vector database like pinecone oder weaviate.

Then I want the program to show in each round two pictures alongside, and you have to chose by clicking the most appealing.

The neural net in the back should then learn based on the previous preference final the best pic.

Say, you have 10.000 and can find with 10 clicks the very best.

I’ve heard from TripleLoss and RankNet.

Are those algorithms suitable? Which one suits best?

How to implement them in PyTorch?

Thanks.

2

but you have no clue, what that would be.
(or how to evaluate it)

again, this is a ‘silly’ question, sorry to say so…

3

nah, that just sounds like he wants to get training data from humans by presenting two choices and making the human choose, and then training the model on that.

or maybe it’s some kind of retrieval task where he expects the “10000” pictures to be indexed by feature vector, and the network discovers a user’s preference function by considering which one of a pair of images the user prefers.

in any case, this has NOTHING to do with OpenCV.