search

gan3sh500 / local-relational-nets

A Pytorch implementation for the paper Local Relational Networks for Image Recognition (https://arxiv.org/pdf/1904.11491.pdf)
112 stars 21 forks source link

Question about the input in "Geometric Prior" #3

Open GongweiChen opened 4 years ago

GongweiChen commented 4 years ago

https://github.com/gan3sh500/local-relational-nets/blob/55eef1d38ae0d9f299fb6f03e893fde141a5c423/layer.py#L9

Thanks for your great and elegant coding! I have a question about "Geometric Prior". Why the inputs are random numbers, while it should be relative position in the original paper?

hust-nj commented 4 years ago

https://github.com/gan3sh500/local-relational-nets/blob/55eef1d38ae0d9f299fb6f03e893fde141a5c423/layer.py#L9

Thanks for your great and elegant coding! I have a question about "Geometric Prior". Why the inputs are random numbers, while it should be relative position in the original paper?

I found the same issue. So wired about this. Do you have any idea?

hust-nj commented 4 years ago

https://github.com/gan3sh500/local-relational-nets/blob/55eef1d38ae0d9f299fb6f03e893fde141a5c423/layer.py#L9

Thanks for your great and elegant coding! I have a question about "Geometric Prior". Why the inputs are random numbers, while it should be relative position in the original paper?

Hi, I find the paper proposes a method to learn geometric prior values denoted as LR(G*) image