M3VSNet
M3VSNet copied to clipboard
whubaichuan
RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation
RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.FloatTensor [1, 3, 3]] is at version 17; expected version 13 instead. Hint: enable anomaly detection to find the operation that failed to compute its gradient, with torch.autograd.set_detect_anomaly(True).
Maybe the codes need to be changed with different variables
I could run it successfully. Do you delete some code such as ".clone()"???
I could run it successfully. Do you delete some code such as ".clone()"???
I don't edit the codes. But I'm wondering if it is relevant to the environment because I don't specifically check my environment to match the required environment.
@agenthong You need to check your environment or google this error.
@whubaichuan Thanks. By the way can you share the requests list hence we can directly install the envireonment.
@agenthong Here
Thanks for replying. I can't directly use ' pip install -r ' to install env.
Can you share a necessary env list in valid format?
Update：I change the pytorch version to match the requirements env then this problem is solved.
Update：I change the pytorch version to match the requirements env then this problem is solved.
I met the same error, could you tell me which version of pytorch to solve this problem? Thanks a lot!
Update：I change the pytorch version to match the requirements env then this problem is solved.
I met the same error, could you tell me which version of pytorch to solve this problem? Thanks a lot!
Actually I didn't solve this question even if I use the corresponding pytorch version. I'm checking the code to figure out if I can fix it.
The error place is mvsnet.py:329
K_xyz_src = torch.matmul(intrinsics_src, xyz_src) #B*3*20480
the intrinsics_src seems to be changed in later
I change the code to
intrinsics_src_ = intrinsics_src
K_xyz_src = torch.matmul(intrinsics_src_, xyz_src) #B*3*20480
it works but i don't know if it's right.
The error place is mvsnet.py:329
K_xyz_src = torch.matmul(intrinsics_src, xyz_src) #B*3*20480the intrinsics_src seems to be changed in later I change the code tointrinsics_src_ = intrinsics_src K_xyz_src = torch.matmul(intrinsics_src_, xyz_src) #B*3*20480it works but i don't know if it's righ
I tried this but it didn't work.
The error place is mvsnet.py:329
K_xyz_src = torch.matmul(intrinsics_src, xyz_src) #B*3*20480the intrinsics_src seems to be changed in later I change the code tointrinsics_src_ = intrinsics_src K_xyz_src = torch.matmul(intrinsics_src_, xyz_src) #B*3*20480it works but i don't know if it's righ
I tried this but it didn't work.
I also change all loss+= to loss = loss + in mvsnet.py. Maybe you can try.
I also encountered the same problem. (In the same environment.) I not only tried using a=a+b instead of a+=b, but also modified the code like this. intrinsics_src_ = intrinsics_src K_xyz_src = torch.matmul(intrinsics_src_, xyz_src) #B320480
It's still not working.
The error message shows as follows:
sys:1: RuntimeWarning: Traceback of forward call that caused the error:
File "train.py", line 398, in
Traceback (most recent call last):
File "train.py", line 398, in
I also encountered the same problem. (In the same environment.) I not only tried using a=a+b instead of a+=b, but also modified the code like this. intrinsics_src_ = intrinsics_src K_xyz_src = torch.matmul(intrinsics_src_, xyz_src) #B_3_20480
It's still not working.
The error message shows as follows:
sys:1: RuntimeWarning: Traceback of forward call that caused the error: File "train.py", line 398, in train() File "train.py", line 155, in train loss,scalar_outputs,image_outputs = train_sample(sample, detailed_summary=do_summary) File "train.py", line 271, in train_sample loss,loss_s,loss_photo,loss_ssim,mask_calculate,mask_num,loss_perceptual,loss_normal,normal_by_depth,error_depth_by_normal,depth_by_normal=model_loss(depth_est,intrinsics,extrinsics,sample_cuda["imgs"],mask_photometric,outputs_feature) File "/home/jojo/Documents/github/M3VSNet/models/mvsnet.py", line 1024, in mvsnet_loss x_src_perceptual,y_src_perceptual=project_with_depth(depth_est_perceptual, ref_intrinsics_perceptual, ref_extrinsics, src_intrinsics_perceptual, src_extrinsics[i]) File "/home/jojo/Documents/github/M3VSNet/models/mvsnet.py", line 320, in project_with_depth K_xyz_src = torch.matmul(intrinsics_src, xyz_src) #B_3_20480
Traceback (most recent call last): File "train.py", line 398, in train() File "train.py", line 155, in train loss,scalar_outputs,image_outputs = train_sample(sample, detailed_summary=do_summary) File "train.py", line 273, in train_sample loss.backward() File "/home/jojo/anaconda3/envs/m3vsnet/lib/python3.6/site-packages/torch/tensor.py", line 102, in backward torch.autograd.backward(self, gradient, retain_graph, create_graph) File "/home/jojo/anaconda3/envs/m3vsnet/lib/python3.6/site-packages/torch/autograd/init.py", line 90, in backward allow_unreachable=True) # allow_unreachable flag RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation
The problem has been solved on my side, by changing the code like this: K_xyz_src = torch.matmul(intrinsics_src.clone(), xyz_src.clone())