godisboy / drpan Goto Github PK

DRPAN produces amazing results, but the ability to save and restore training checkpoints would be really helpful for training on a single GPU, especially on large datasets.

I use your code to dehaze,but there are many dots on the picture,when you train in deraining task,can you
meet this issue,and what may cause it?thanks

When I try to run main.py, an error occurs:
ImportError: dynamic module does not define module export function (PyInit__crop_and_resize)

raceback (most recent call last): File "main.py", line 9, in <module> from src.trainer import trainer_gan File "/home/leander/linli/code/DRPAN/DRPAN/src/trainer.py", line 11, in <module> from roi_align.roi_align import RoIAlign File "/home/leander/linli/code/DRPAN/DRPAN/roi_align/roi_align.py", line 4, in <module> from .crop_and_resize import CropAndResizeFunction, CropAndResize File "/home/leander/linli/code/DRPAN/DRPAN/roi_align/crop_and_resize.py", line 7, in <module> from ._ext import crop_and_resize as _backend File "/home/leander/linli/code/DRPAN/DRPAN/roi_align/_ext/crop_and_resize/__init__.py", line 3, in <module> from ._crop_and_resize import lib as _lib, ffi as _ffi ImportError: dynamic module does not define module export function (PyInit__crop_and_resize)

I don't know how to solve this problem, can you help?
Thank you.

Hello sir.
I have interesting on your model.
So i want to try training DRPAN.
But i met some problems.
My environment : cuda 11.x , pytorch 1.4, python 3.7
Can i try training your model ?
If possible, please tell me how to try.

Thanks.

Hi，I am very interested in this work，can you provide the test model，thank you very much.

Thank you very much for the contribution and excellent work!

DRPAN it's an excellent idea that I'd like to build upon. However, I am having problems to reproduce the work.
Is there a list of requirements (preferably conda environment or dockerfile) to make sure my system is correctly set up?

Using any pytorch version newer than 0.4.1 there seems to be incompatibility issues due to some deprecated functions. However when using 0.4.1 things seem to work but when launching training running main.py (for any DRPAN, Pix2PixHD_DRPAN or cycleGAN_DRPA) the following happens:

Traceback (most recent call last):
File "DRPAN/DRPAN/main.py", line 9, in
from src.trainer import trainer_gan
File "/home/jupyter/DRPAN/DRPAN/src/trainer.py", line 11, in
from roi_align.roi_align import RoIAlign
File "/home/jupyter/DRPAN/DRPAN/roi_align/roi_align.py", line 4, in
from .crop_and_resize import CropAndResizeFunction, CropAndResize
File "/home/jupyter/DRPAN/DRPAN/roi_align/crop_and_resize.py", line 7, in
from ._ext import crop_and_resize as _backend
File "/home/jupyter/DRPAN/DRPAN/roi_align/_ext/crop_and_resize/init.py", line 3, in
from ._crop_and_resize import lib as _lib, ffi as _ffi
ImportError: /home/jupyter/DRPAN/DRPAN/roi_align/_ext/crop_and_resize/_crop_and_resize.so: undefined symbol: state

How can I make sure my environment is the correctly set to reproduce your work?
Thank you so much!

I would like to try the architecture but at the time I launch the training, as indicated, I have the following error message:
  from ._crop_and_resize import lib as _lib, ffi as _ffi
ImportError: dynamic module does not define module export function (PyInit__crop_and_resize).
I looked for solutions but I didn't find anything useful.
I use archlinux as os, I tried both with python 3.7 and 3.6 but the error is the same, any suggestions?
Thanks.

Hi, thanks a lot for your awesome work.
I found the FCN-8s score of Pix2pix in https://arxiv.org/pdf/1611.07004v1.pdf is the same as the score of Pix2pix shown in your paper. Pix2pix uses 256x256 resolution images to calculate FCN-8s score, but the paper shows the result of 512x512. Can you tell me what resolution you use to calculate the FCN-8s score