Open lina926 opened 2 years ago
We need more infomation.
the config : rotated_retinanet_obb_kld_stable_r50_fpn_1x_dota_le90.py it can run successful but when i debug, the Sigma_t(kld_loss) is : tensor([[[2.5000e-15, 0.0000e+00], [0.0000e+00, 2.5000e-15]],
[[2.5000e-15, 0.0000e+00],
[0.0000e+00, 2.5000e-15]],
[[2.5000e-15, 0.0000e+00],
[0.0000e+00, 2.5000e-15]],
...,
[[2.5000e-15, 0.0000e+00],
[0.0000e+00, 2.5000e-15]],
[[2.5000e-15, 0.0000e+00],
[0.0000e+00, 2.5000e-15]],
[[2.5000e-15, 0.0000e+00],
[0.0000e+00, 2.5000e-15]]], device='cuda:0')
I just want to know why the target information is 0
can you tell me the difference between GD_loss and GD_loss_v1? when i debug, i find the input target of kld_loss in GD_loss is 0 such as (0,0,0,0,0,), why?