How to understand the idea of the algorithm in the file "auction_match_gpu.cu"?

[ChambinLee]

Hello. I have read the code in the file "auction_match_gpu.cu" for several days, and I try to understand how you calculate the EMD. I have some trouble on reading it. Could you give me some hints about how to understand these codes or paper about this? Thank you in advance!

[ZirongLiu]

I also encountered the same problem, I wanted to upsample more points, such as 8192, but I couldn't get started.

[lyqun]

I also encountered the same problem, I wanted to upsample more points, such as 8192, but I couldn't get started.

The EMD (https://github.com/yulequan/PU-Net/tree/master/code/tf_ops/emd) only supports processing with at most 4,096 points. Here, I suggest 1.) reduce the number of points; 2.) remove the EMD loss (use CD only); 3.) randomly sample 4,096 points from upsampled 8,192 points for EMD loss calculation; or 4.) modify the source code to support the processing of more points.