python testcuda.py error

[xtanitfy]

(Pytorch-1.4.0) sh-4.3$python testcuda.py
torch.Size([2, 64, 128, 128])
torch.Size([20, 32, 7, 7])
torch.Size([20, 32, 7, 7])
torch.Size([20, 32, 7, 7])
0.971507, 1.943014
0.971507, 1.943014
Zero offset passed
/home/ma-user/anaconda3/envs/Pytorch-1.4.0/lib/python3.6/site-packages/torch/autograd/gradcheck.py:302: UserWarning: The {}th input requires gradient and is not a double precision floating point or complex. This check will likely fail if all the inputs are not of double precision floating point or complex.
'The {}th input requires gradient and '
check_gradient_dpooling: True
Traceback (most recent call last):
File "testcuda.py", line 265, in
check_gradient_dconv()
File "testcuda.py", line 97, in check_gradient_dconv
eps=1e-3, atol=1e-4, rtol=1e-2))
File "/home/ma-user/anaconda3/envs/Pytorch-1.4.0/lib/python3.6/site-packages/torch/autograd/gradcheck.py", line 390, in gradcheck
checkIfNumericalAnalyticAreClose(a, n, j)
File "/home/ma-user/anaconda3/envs/Pytorch-1.4.0/lib/python3.6/site-packages/torch/autograd/gradcheck.py", line 372, in checkIfNumericalAnalyticAreClose
'numerical:%s\nanalytical:%s\n' % (i, j, n, a))
File "/home/ma-user/anaconda3/envs/Pytorch-1.4.0/lib/python3.6/site-packages/torch/autograd/gradcheck.py", line 289, in fail_test
raise RuntimeError(msg)
RuntimeError: Jacobian mismatch for output 0 with respect to input 1,
numerical:tensor([[ 0.0000, 0.0000, 0.0000, ..., 0.0000, 0.0000, 0.0000],
[ 0.0000, 0.0041, 0.0000, ..., 0.0000, 0.0000, 0.0000],
[ 0.0000, 0.0000, 0.0000, ..., 0.0000, 0.0000, 0.0000],
...,
[ 0.0000, 0.0000, 0.0000, ..., -0.0018, 0.0000, 0.0000],
[ 0.0000, 0.0000, 0.0000, ..., 0.0000, 0.0000, 0.0000],
[ 0.0000, 0.0000, 0.0000, ..., 0.0000, 0.0000, 0.0009]],
device='cuda:0')
analytical:tensor([[ 0.0000, 0.0000, 0.0000, ..., 0.0000, 0.0000, 0.0000],
[ 0.0000, 0.0041, 0.0000, ..., 0.0000, 0.0000, 0.0000],
[ 0.0000, 0.0000, 0.0000, ..., 0.0000, 0.0000, 0.0000],
...,
[ 0.0000, 0.0000, 0.0000, ..., -0.0018, 0.0000, 0.0000],
[ 0.0000, 0.0000, 0.0000, ..., 0.0000, 0.0000, 0.0000],
[ 0.0000, 0.0000, 0.0000, ..., 0.0000, 0.0000, 0.0009]],
device='cuda:0')

[xtanitfy]

chage the line 97 as follow and the error dispear:

print('check_gradient_dconv: ',
gradcheck(dcn_v2_conv, (input, offset, mask, weight, bias,
stride, padding, dilation, deformable_groups),
eps=1e-3, atol=1e-3, rtol=1e-2))

[lucasjinreal]

I am not sure this version compatible with pytorcch 1.4, it is tested on latest version of pytorch

[peizhaoli05]

I am having the same error while using PyTorch 1.8.1/1.7.1.