Training problems

[Edric-star]

Hi, I confronted this problem while doing 4.1.2 training, are there any other libraries or packages that you could offer to help?Thank you so much! I switched my conda env to pasco while doing this...

from uncertainty.models.metrics import SSCMetrics

ModuleNotFoundError: No module named 'uncertainty'

[anhquancao]

Thanks for reporting the bug. Sorry about the name error. I've fixed it and pushed the update. Please try again.

[Edric-star]

I tried pip install uncertainty and it turned out(I'm not pretty sure) a fail and I supposed it might be the problem of the version of numpy because my numpy is 1.25.2 which's not the same as the numpy==1.20 in requirements.txt and if I did pip intall requirements.txt, it would reported as follows:

The conflict is caused by:
The user requested numpy==1.20
h5py 3.7.0 depends on numpy>=1.14.5
imageio 2.33.1 depends on numpy
numba 0.58.1 depends on numpy<1.27 and >=1.22

Then I set a numpy==1.25 to fit this problem and I guess it might be a possible mistake later, but I'm really confused.Professor Cao, could you please review your requirements.txt and check the 'uncertainty' module? I really appreciate your work and that's why I started running the model as soon as you updated these days.

[Edric-star]

Thanks for reporting the bug. Sorry about the name error. I've fixed it and pushed the update. Please try again.

I see and thank you for your immediate replay！I would try again soon

[anhquancao]

You should install a lower version of numba to be able to install NumPy 1.20. Recent NumPy versions use floatinstead of np.float, which causes problems with the code.

[Edric-star]

Collaborator Thanks, I got it fixed. By the way, would you decide to release the eval code that's able to generate the completed label? I mean the Panoptic scene completion.

[anhquancao]

Hi, you can find the code for generating the panoptic label here.

[Edric-star]

I did notice that later and that's why I closed the comments, anyway thanks!