search

microsoft / computervision-recipes

Best Practices, code samples, and documentation for Computer Vision.
MIT License
9.56k stars 1.18k forks source link

[ASK] Batch Sampler and extreme classification #662

Closed kifjj closed 3 years ago

kifjj commented 3 years ago

Description

Hi @PatrickBue thanks for the similarity source code, very useful to see the implementation of such an important paper. I have a question, the paper mentions Subsampling for Classification Scalability and in the case of the SOP Dataset I think we are in extreme classification situation. You didn't implement that and still replicate the performance, do you have any explanation? Do you think that implementing subsampling would be beneficial for other datasets with 27k classes?

Thanks #

Other Comments

PatrickBue commented 3 years ago

Hi, I don't fully remember the details of the paper hence don't know the answer. Maybe the original authors of the paper/code would know: https://github.com/azgo14/classification_metric_learning

kifjj commented 3 years ago

Thanks for the reply @PatrickBue.