Can it run directly?!

[hiCplusplus1900]

Greetings! First of all, thank you for your excellent work!

Regarding the code you posted, did you actually test it? Does it run straight away? There are too many errors in the code, not limited to filename reference errors, CUDA interface with pytorch triggered errors, etc. I have strictly followed the cuda11.3 and pytorch1.10.0 environments you mentioned, but there are still a lot of problems. Also, your code doesn't support GPU parallelism, so why write it in your code? It's really so incomprehensible. Is this the quality of code released for AAAI papers? Did you release such code on purpose or was it accidental? Your inadvertent action makes me question whether the code you released can get close to the results of your paper!

Could you explain something?

[liujiuming123]

Firstly, thank you for your suggestion!
We have checked our codes. Actually, there is only one filename error in line 18 of train.py, where 'from translo_model import ...' is miswritten as 'from translonet_model import ...' due to the different code version. We have noticed this error and updated it now for your kindly advice. Also, we are still open to any other question about the errors you mentioned when running our codes. However, we strongly advise you can provide the detailed problems you encountered with figure. In this way, we can help you to fix them.

If you can't configure the environment yourself, we will also provide you with our docker environment to replicate the results in paper.

[hiCplusplus1900]

Thanks for your reply.

I have solved the problems and run your code. However, during the training process, the convergence speed of the model seemed to be a little slow. I remember running for two or three days, but the result didn't seem to drop. How many epochs does your model need to run to achieve the result in the paper? Post my running results below.

[YLRK]

May I ask you how to solve the problem in the code? I run the code according to the Settings, and I encounter the following problem.

CUDA error: device-side assert triggered
File "/home/user/Projects/TransLO/conv_util.py", line 290, in forward
qi_xyz_grouped = torch.gather(xyz2_bn3, 1, neighbor_idx.unsqueeze(-1).repeat(1, 1, 3))
File "/home/user/Projects/TransLO/translo_model.py", line 355, in forward
l2_cost_volume_origin = self.cost_volume1(l2_xyz_proj_f1, l2_xyz_proj_f2, l2_points_cross_proj_f1, l2_points_cross_proj_f2)
File "/home/user/Projects/TransLO/train.py", line 157, in main
l0_q, l0_t, l1_q, l1_t, l2_q, l2_t, l3_q, l3_t, pc1_ouput, q_gt, t_gt, w_x, w_q = model(pos2, pos1, T_gt, T_trans, T_trans_inv)
File "/home/user/Projects/TransLO/train.py", line 305, in
main()
RuntimeError: CUDA error: device-side assert triggered

[hiCplusplus1900]

Hi, @YLRK Sorry to reply so late.
I had a problem like yours, but I forgot how to fix it. I compared my code with the author's, and I mainly modified the following aspects, just for reference.
The first is to change the data type Long in the C++ interface to Int. The corresponding Python reference section also needs to be modified.
Second, modify the code "loaded on cuda" in train.py, just as you usually use it, and comment out its original part.

[liujiuming123]

The training process is about 2~3 days. It takes 300 epochs for convergence, and experiment results will be saved in file /logs/output.xlsx. Our previous results are also given below.

[liujiuming123]

May I ask you how to solve the problem in the code? I run the code according to the Settings, and I encounter the following problem.

CUDA error: device-side assert triggered
File "/home/user/Projects/TransLO/conv_util.py", line 290, in forward
qi_xyz_grouped = torch.gather(xyz2_bn3, 1, neighbor_idx.unsqueeze(-1).repeat(1, 1, 3))
File "/home/user/Projects/TransLO/translo_model.py", line 355, in forward
l2_cost_volume_origin = self.cost_volume1(l2_xyz_proj_f1, l2_xyz_proj_f2, l2_points_cross_proj_f1, l2_points_cross_proj_f2)
File "/home/user/Projects/TransLO/train.py", line 157, in main
l0_q, l0_t, l1_q, l1_t, l2_q, l2_t, l3_q, l3_t, pc1_ouput, q_gt, t_gt, w_x, w_q = model(pos2, pos1, T_gt, T_trans, T_trans_inv)
File "/home/user/Projects/TransLO/train.py", line 305, in
main()
RuntimeError: CUDA error: device-side assert triggered

Sorry, I have not met this error before. Maybe it is caused by different CUDA versions, or array overreach issue?

[rasrab1992]

May I ask you how to solve the problem in the code? I run the code according to the Settings, and I encounter the following problem.

CUDA error: device-side assert triggered
File "/home/user/Projects/TransLO/conv_util.py", line 290, in forward
qi_xyz_grouped = torch.gather(xyz2_bn3, 1, neighbor_idx.unsqueeze(-1).repeat(1, 1, 3))
File "/home/user/Projects/TransLO/translo_model.py", line 355, in forward
l2_cost_volume_origin = self.cost_volume1(l2_xyz_proj_f1, l2_xyz_proj_f2, l2_points_cross_proj_f1, l2_points_cross_proj_f2)
File "/home/user/Projects/TransLO/train.py", line 157, in main
l0_q, l0_t, l1_q, l1_t, l2_q, l2_t, l3_q, l3_t, pc1_ouput, q_gt, t_gt, w_x, w_q = model(pos2, pos1, T_gt, T_trans, T_trans_inv)
File "/home/user/Projects/TransLO/train.py", line 305, in
main()
RuntimeError: CUDA error: device-side assert triggered

Sorry, I have not met this error before. Maybe it is caused by different CUDA versions, or array overreach issue?

Hello, I also had this error, although I tried to use docker with the same cuda and python version that you have mentioned. Could you please check it again and give more detail about python version, cuda version, pytorch version, cudnn, .... . Or maybe it is better to share your docker file or image.

Best regards

[rasrab1992]

h.gather(xyz2_bn3, 1, neighbor_idx.unsqueeze(-1).repeat(1, 1, 3))
File "/home/user/Projects/TransLO/translo_model.py", line 355, in forward
l2_cost_volume_origin = self.cost_volume1(l2_xyz_proj_f1, l2_xyz_proj_f2, l2_points_cross_proj_f1, l2_points_cross_proj_f2)
File "/home/user/Projects/TransLO/train.py", line 157, in main
l0_q, l0_t, l1_q, l1_t, l2_q, l2_t, l3_q, l3_t, pc1_ouput, q_gt, t_gt, w_x, w_q = model(pos2, pos1, T_gt, T_trans, T_trans_inv)
File "/home/user/Projects/TransLO/train.py", line 305, in

Hello @YLRK , I had the same problem. Did you solve it?

[YLRK]

Hi @rasrab1992 , I'm sorry, I have been busy with other projects recently, so I have no time to run this work, and this problem has not been solved for the time being, Maybe you can ask @hiCplusplus1900 for help.

[rasrab1992]

Okay Thanks. Good luck @YLRK.

Hi @rasrab1992 , I'm sorry, I have been busy with other projects recently, so I have no time to run this work, and this problem has not been solved for the time being, Maybe you can ask @hiCplusplus1900 for help.

[rasrab1992]

@hiCplusplus1900

[liujiuming123]

May I ask you how to solve the problem in the code? I run the code according to the Settings, and I encounter the following problem.

CUDA error: device-side assert triggered
File "/home/user/Projects/TransLO/conv_util.py", line 290, in forward
qi_xyz_grouped = torch.gather(xyz2_bn3, 1, neighbor_idx.unsqueeze(-1).repeat(1, 1, 3))
File "/home/user/Projects/TransLO/translo_model.py", line 355, in forward
l2_cost_volume_origin = self.cost_volume1(l2_xyz_proj_f1, l2_xyz_proj_f2, l2_points_cross_proj_f1, l2_points_cross_proj_f2)
File "/home/user/Projects/TransLO/train.py", line 157, in main
l0_q, l0_t, l1_q, l1_t, l2_q, l2_t, l3_q, l3_t, pc1_ouput, q_gt, t_gt, w_x, w_q = model(pos2, pos1, T_gt, T_trans, T_trans_inv)
File "/home/user/Projects/TransLO/train.py", line 305, in
main()
RuntimeError: CUDA error: device-side assert triggered

Sorry, I have not met this error before. Maybe it is caused by different CUDA versions, or array overreach issue?

Hello, I also had this error, although I tried to use docker with the same cuda and python version that you have mentioned. Could you please check it again and give more detail about python version, cuda version, pytorch version, cudnn, .... . Or maybe it is better to share your docker file or image.

Best regards

Thank you for your question! I have checked our environment again. It is based on CUDA 11.3.1, pytorch 1.10.0. We will release our docker here.

[Dlemand]

May I ask you how to solve the problem in the code? I run the code according to the Settings, and I encounter the following problem.

CUDA error: device-side assert triggered
File "/home/user/Projects/TransLO/conv_util.py", line 290, in forward
qi_xyz_grouped = torch.gather(xyz2_bn3, 1, neighbor_idx.unsqueeze(-1).repeat(1, 1, 3))
File "/home/user/Projects/TransLO/translo_model.py", line 355, in forward
l2_cost_volume_origin = self.cost_volume1(l2_xyz_proj_f1, l2_xyz_proj_f2, l2_points_cross_proj_f1, l2_points_cross_proj_f2)
File "/home/user/Projects/TransLO/train.py", line 157, in main
l0_q, l0_t, l1_q, l1_t, l2_q, l2_t, l3_q, l3_t, pc1_ouput, q_gt, t_gt, w_x, w_q = model(pos2, pos1, T_gt, T_trans, T_trans_inv)
File "/home/user/Projects/TransLO/train.py", line 305, in
main()
RuntimeError: CUDA error: device-side assert triggered

Sorry, I have not met this error before. Maybe it is caused by different CUDA versions, or array overreach issue?

Hello, I also had this error, although I tried to use docker with the same cuda and python version that you have mentioned. Could you please check it again and give more detail about python version, cuda version, pytorch version, cudnn, .... . Or maybe it is better to share your docker file or image.
Best regards

Thank you for your question! I have checked our environment again. It is based on CUDA 11.3.1, pytorch 1.10.0. We will release our docker here.

Looking forward to your docker environment！

[rasrab1992]

Hello @liujiuming123,
Still looking forward to your docker environment!!! Or at least some details about Python version, ubuntu version, driver .....
Is it difficult? Maybe you don't prefer to other guys site your paper.

Thanks.

[liujiuming123]

Hi! We prepared CVPR last few weeks, so did not upload the environment. Sorry for that. However, the former docker installed lots of packages and the memory exceeds 100G. It is too large for delivering and we are trying to compress it recently. Once it processed over, we will upload as soon as possible. Best regards, jiuming

…

----- 原始邮件 ----- 发件人: "Rasool_Rabee_1992" ***@***.***> 收件人: "IRMVLab/TransLO" ***@***.***> 抄送: "liujiuming123" ***@***.***>, "Mention" ***@***.***> 发送时间: 星期二, 2023年 12 月 12日 上午 1:56:07 主题: Re: [IRMVLab/TransLO] Can it run directly?! (Issue #2) Hello @liujiuming123, Still looking forward to your docker environment!!! Or at least some details about Python version, ubuntu version, driver ..... Is it difficult? Maybe you don't prefer to other guys site your paper. Thanks.

-- Reply to this email directly or view it on GitHub: #2 (comment) You are receiving this because you were mentioned. Message ID: ***@***.***>