Hello author, thank you for your excellent work. I encountered some problems while reproducing your work, and I hope you can take some time out of your busy schedule to answer them. Thank you!
- I see that you said: [NATTEN] Please refer to https://shi-labs.com/natten to install NATTEN with correct CUDA and PyTorch versions (Note: we trained the model using CUDA 12.1 + PyTorch 2.2, and NATTEN=0.15.1). For example, we can install NATTEN with Pytorch 2.2 and CUDA 12.1 with
After successfully installing this package and PyTorch 2.2 Cuda12.1, and then successfully installing tvdcn, running the program has been unsuccessful with a prompt:
RuntimeError: Couldn't load custom C++ ops. Recompile C++ extension with:python setup.py build_ext --inplace
Then I followed the instructions of the tvdcn library, but it was still unsuccessful. After multiple attempts, I found that it should be related to the version of PyTorch. It seems that this library does not support 2.2, but supports 2.1. So I would like to consult the author. If you are using 2.2, have you encountered this problem and how have you handled it?
-
Unable to reproduce the results of the paper. I have conducted multiple experiments on the rented cloud server, all of which are Tiny versions, as follows:
(1) Python 3.8 Pytorch 2.2 Cuda 12.1, NATTEN=0.15.1, V100, deformable convolution implemented using torch: mean-dice: 0.828499 mean-hd95: 24.450771
(2) Python 3.8 Pytorch 2.1 Cuda12.1, NATTEN=0.15.13090, using tvdcn: mean-dice: 0.819928 mean-hd95: 24.864291
(3) Python 3.8 Pytorch 2.1 Cuda12.1, NATTEN=0.15.1, V100, using tvdcn: mean-dice: 0.827410 mean-hd95: 20.259738
-
What is the Python version of the author and would like to keep it consistent with yours
-
I want to use weights to see if it's the influence of the environment, but it seems like there's no weight for Tiny w DS
The above is my question. Thank you for answering it. Could you please help me? This is very important to me. I would like to reproduce your paper and cite it for you. Thank you again, author.
Hello author, thank you for your excellent work. I encountered some problems while reproducing your work, and I hope you can take some time out of your busy schedule to answer them. Thank you!
After successfully installing this package and PyTorch 2.2 Cuda12.1, and then successfully installing tvdcn, running the program has been unsuccessful with a prompt:
RuntimeError: Couldn't load custom C++ ops. Recompile C++ extension with:python setup.py build_ext --inplace
Then I followed the instructions of the tvdcn library, but it was still unsuccessful. After multiple attempts, I found that it should be related to the version of PyTorch. It seems that this library does not support 2.2, but supports 2.1. So I would like to consult the author. If you are using 2.2, have you encountered this problem and how have you handled it?
Unable to reproduce the results of the paper. I have conducted multiple experiments on the rented cloud server, all of which are Tiny versions, as follows:
(1) Python 3.8 Pytorch 2.2 Cuda 12.1, NATTEN=0.15.1, V100, deformable convolution implemented using torch: mean-dice: 0.828499 mean-hd95: 24.450771
(2) Python 3.8 Pytorch 2.1 Cuda12.1, NATTEN=0.15.13090, using tvdcn: mean-dice: 0.819928 mean-hd95: 24.864291
(3) Python 3.8 Pytorch 2.1 Cuda12.1, NATTEN=0.15.1, V100, using tvdcn: mean-dice: 0.827410 mean-hd95: 20.259738
What is the Python version of the author and would like to keep it consistent with yours
I want to use weights to see if it's the influence of the environment, but it seems like there's no weight for Tiny w DS
The above is my question. Thank you for answering it. Could you please help me? This is very important to me. I would like to reproduce your paper and cite it for you. Thank you again, author.