Giters

google-research / reassessed-imagenet

Labels and other data for the paper "Are we done with ImageNet?"

Home Page:https://arxiv.org/abs/2006.07159

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Indexing

ukoehler opened this issue · comments

commented

Hi,

I have a problem relating the information in real.json to the images and classes. The first problem seems to be in numpy.py line 12. I guess that should be i+1 instead of i+i?

Second, when I look at image ILSVRC2012_val_00000140.JPEG (a guy playing a flute with a dog) and read the 140th entry in real.json I get 163. Following the file meta_clsloc.mat provided with the ImageNet data that would be American_black_bear? That cannot be right.

Lets examine the image before and after: no, black bear is wrong for ILSVRC2012_val_00000139.JPEG and ILSVRC2012_val_00000141.JPEG, as well. Try category 162 water_buffalo and 164 Angora do not fit the image in question nor the neighbor images.

Where am I going wrong?