Geometric verification is a pairwise method for comparison content on two images based on classical descriptors. It's especially good in landmark detection task and was widely used on Kaggle in the relevant competitions. You can easily find more details in the internet.
From OML's point of view, we can implement this verifier and pass it to the existing PairwiseReranker object. See similar code for Siamese model which compares two images on pixel level. (You only need to check and adapt the prediction code, because geometric verifier is not a trainable NN, it's rather an algorithm).
AlekseySh
commented
2 weeks ago
Geometric verification is a pairwise method for comparison content on two images based on classical descriptors. It's especially good in landmark detection task and was widely used on Kaggle in the relevant competitions. You can easily find more details in the internet.
From OML's point of view, we can implement this verifier and pass it to the existing
PairwiseRerankerobject. See similar code for Siamese model which compares two images on pixel level. (You only need to check and adapt the prediction code, because geometric verifier is not a trainable NN, it's rather an algorithm).