search

quziyan / NPairLoss

MPI-CAFFE implementation of <Improved Deep Metric Learning with Multi-class N-pair Loss Objective>
11 stars 7 forks source link

A question about N Pair Loss #1

Open FanQC opened 6 years ago

FanQC commented 6 years ago

Hi,

Thank u for your implementation of n-pair loss. It helps me a lot. But, I have a question about the operation before computing the loss. Would u mind giving me a expaination ?

My e-mail: fanqcamoy@163.com

quziyan commented 6 years ago

This implementation is the npairloss with online negative pairs mining based on statistic within a big-batch. I think the difficulty to understand the code is refer to CUDA.

FanQC commented 6 years ago

Thank u for replying me. Actually, I was confused about the online negative pairs mining based on statistic. Would u mind giving me some paper or web pages about it.Thanks a lot.

quziyan commented 6 years ago

FaceNet: A Unified Embedding for Face Recognition and Clustering Improved Deep Metric Learning with Multi-class N-pair Loss Objective

you could find some basic instruction in the negative data mining section

FanQC commented 6 years ago

Thanks a lot.