RuntimeError: PyTorch was compiled without NumPy support

[elkizana]

I tried with NumPy 1.18 then downgraded to 1.15 with the same error message.

[MingSun-Tse]

I sense your problem may be that you didn't install the pytorch the right way. You may want to look at the standard installation guide and especially, use conda to manage the environment. https://pytorch.org/get-started/locally/

[elkizana]

When I comment line 128 in WCT.py no error is giving. This is the full error output : File "WCT.py", line 128, in <module> vutils.save_image(cImg.data.cpu(), out_path) File "/usr/local/lib/python3.7/site-packages/torchvision/utils.py", line 102, in save_image ndarr = grid.mul(255).clamp(0, 255).byte().permute(1, 2, 0).cpu().numpy() RuntimeError: PyTorch was compiled without NumPy support

[MingSun-Tse]

Oh, that's interesting. But that functionality is still needed since you need to save stylized images. Could you please give your torchvision version?

[elkizana]

torch 0.4.1 torchvision 0.2.1

Please after looking to this problem I would like to stylize my own picture, what's the correct command I should provide. Thanks a lot for you efforts MingSun-Tse !

[MingSun-Tse]

Hi, for now I don't know exactly what the problem is either. I can run on my computer normally based on what the README provides. To avoid potential environment issues, I just pushed the requirements.txt to the repo. I am sorry but you may need to rebuild your environment from scratch and try again. After that, if the problem remains, please let me know. Thanks!

[elkizana]

Strange. I uninstalled all the packages and reinstalled them from scratch from the requirements.txt file but didn't get rid of the error. This repo seems to be the only solution I found so far, so that I can achieve my purpose of transferring style to a large sized image I have.

[elkizana]

MingSun-Tse How much gpu and ram resources are needed to treat a 10k pixel image ? Thanks.

[MingSun-Tse]

Hi Barbarosse -- I don't remember the exact numbers. I will run again and let you know after. Thanks. By the way, did you resolve the weird numpy problem?

[elkizana]

Hi, no not yet. But I managed to run it in Colab tough, but the results weren't satisfying for me, compared to other NST codes I used, the result was too off.

[MingSun-Tse]

Could you please report your environment (all used library versions) so that I can reproduce your error?

[elkizana]

OS : Ubuntu 20.04 Python : 3.7.6 All dependencies from your requirement.txt file. Nvidia Driver : 440 Cuda : 10.1

[MingSun-Tse]

Please try python=3.6.9 https://github.com/MingSun-Tse/Collaborative-Distillation#environment I don't know why, but it seems 3.7.6 does not work.

[elkizana]

OK