Imagenet labels incorrect

vision
1

I am using resnet50 on imagenet 2017 dataset from here: http://image-net.org/challenges/LSVRC/2017/
It seems that the “target”, i.e. the labels, are numbered from 1 to 1000. Are we supposed to convert the labels so that they lie in the range [0,999]? If that is the case, do we simply subtract 1 from the original labels, i.e. by using the following command,

  target = (target - torch.ones(bsz).type(torch.LongTensor)).cuda()

? Thanks

2

yes. you can also do:

target = target - 1
3

Hi Smith,
thanks for a quick reply…so is it safe to assume that the labels (targets) are only linearly shifted by a constant amount, i.e., target_ideal = target_actual - 1? And also, it is the same for train and validation sets?
Best,
Rahul