why need norm = np.ones((pred.shape[0], 2)) * np.array([h, w]) / 10 in accuracy about udp-pose HOT 4 CLOSED

@azuryl emm, i think the accuracy is meaningless. So i haven't paid attention on this function. You can question this in the origin hrnet repo.

from udp-pose.

Dear JunJie thank your reply

in

lif config.MODEL.TARGET_TYPE == 'offset':
_, avg_acc, cnt, pred = accuracy(output.cpu().numpy()[:, ::3, :, :],
target.cpu().numpy()[:, ::3,:, :])
you used this function. I do not know why

Thank you

from udp-pose.

@azuryl i just imitate the source hrnet repo to add this for preventing the error.

from udp-pose.

ok I see thank :)

from udp-pose.