Centernet: How to calc dataset specific eigenvalues and eigenvectors

[dvbvr]

Hi,
I'm using CenterNet with my own dataset but I don't really understand the eigenvalues (_eig_val) and eigenvectors (_eig_vec) specified in coco.py. How are they calculated? In the dla repo I saw that they called it the "image pixel variations", do you guys know what that means?
Thank you!

[xingyizhou]

I didn't understand either. It is used for color augmentation. I always use the default eigenvectors as in COCO for other datasets and it works fine.

[evitself]

hey guys, @dvbvr @xingyizhou , I guess this method comes from the paper by Alex and Hinton:
https://papers.nips.cc/paper/4824-imagenet-classification-with-deep-convolutional-neural-networks.pdf