something about the codes of pytorch-spn
Schizophreni opened this issue · 3 comments
There are so many bugs if trying to compile pytorch-spn in another computer with different cuda installation path. your build.py has set the cuda path as '/usr/local/cuda'， which can not easily be spotted. There are other bugs due to the link of library and the installation path of CUDA, can you upload the already compiled library?
I've compiled the pytorch-spn either on windows10+cuda10.0 and ubuntu16.04+cuda9.0, all in vain. I would rather reading your paper for 10hrs rather than debugging your codes for 10+hr. But the reality is I read your paper for 2hrs and get stuck in your codes for 10+hrs just to compile the damed pytorch-spn......
First of all, if you really have spent 2/10 hours on the paper, then you will know pytorch-spn is a prior work and I didn't write it.
Second, it works fine and here is an example:
Third, this is Github issue for issues and not a shrink's sofa for your feelings. The whole paragraph you wrote doesn't provide any useful information for me to help you debug.
Last, please do not use our code, there are other photo-realistic methods and you can insult other authors.
FYI, please learn how to curse correctly, you spelled damned wrong.
Thanks for you reply, which makes me realize I'm still a green hand or little child in front of you. My impulsive behaviour has wasted your valuable time and energy, also has done harm to myself. So I still want to clarify several things:
First of all, I'm sorry for the message that I have sent to you, for the useless words without a tiny useful advice or distinct idea which wasted your precious time and energy. What's more, I shouldn't insult on you.
Second, I should not blame my own fault on you, and I ought to respect other peoples' work. But it turns out that I have done neither. What's more, I should be aware of the consequences of what I say and what I convey. The Github is a community for the coders and researchers to communicate, and exchange ideas, propose ideas and advices, which is very important for the development of the community. I must not share my individual feelings on the community, on the contrary, I need to shed my opinions, sparkle my ideas on this comminity. Thanks for your sincere replication to make me realize this point.
Third, I am really really really appreciated by your work in this paper. In my shallow view, the transformation matrix T has linked the content and style in this task, however, I think it can do much more things. It acts like a domain "fusion" or domain transfer, what may be the key for other hard tasks like: learning to see in the dark, image de-noise etc. So after reading your paper, I cannot wait to do some experiments on the Photo-realistic style transfer, to find out whether the transformation matrix can "work" in other tasks, if so, I can dive more deeper and exploit what is behind this simple linear transformation when it performs better. In another way, I think "attention" is also in it, cause the transformation matrix can learn to leverage the weights of different feature maps. What's more, such a configuration is easy to be implemented and trained.
Fourth, I really really know that pytorch-spn is not the ingredient of your paper. In your paper, SPN is used for anti-distortion in Photo-realistic style transfer tasks. You also realease the link to the original repository of pytorch-spn, so I should not blame the compiling bugs on your work. But I find there is something different between your pytorch-spn and the orginal. I'm gonna to deal with it by myself and seeking help from the original author of pytorch-spn.
Last by not least, thanks for pointing out my spelling mistake, wish you the best
Looking forward to your future work!
Thanks for the clarifications, please feel free to report any issues in the future.