xiaoyu258 / docproj Goto Github PK

Hi!

Do you plan you open up the source code before conference?

Thanks!

Hi,

Can you also provide the Linux binary for GraphCut.exe?

we face issue in executing: resampling.py
my numba lib version is 0.51.2
OS: Windows 10
Python 3.7

Error :
Traceback (most recent call last):
File "resampling.py", line 203, in
resImg, resMsk = rectification(distortedImg, flow)
File "resampling.py", line 194, in rectification
iterSearch[blockspergrid, threadsperblock](padu, padv, paddistorted, resultImg, maxIter, precision, resultMsk)
File "C:\Users\testing\AppData\Local\Programs\Python\Python37\lib\site-packages\numba\cuda\compiler.py", line 770, in call
self.stream, self.sharedmem)
File "C:\Users\testing\AppData\Local\Programs\Python\Python37\lib\site-packages\numba\cuda\compiler.py", line 861, in call
kernel = self.compile(argtypes)
File "C:\Users\testing\AppData\Local\Programs\Python\Python37\lib\site-packages\numba\cuda\compiler.py", line 935, in compile
kernel.bind()
File "C:\Users\testing\AppData\Local\Programs\Python\Python37\lib\site-packages\numba\cuda\compiler.py", line 576, in bind
self.func.get()
File "C:\Users\testing\AppData\Local\Programs\Python\Python37\lib\site-packages\numba\cuda\compiler.py", line 446, in get
ptx = self.ptx.get()
File "C:\Users\testing\AppData\Local\Programs\Python\Python37\lib\site-packages\numba\cuda\compiler.py", line 414, in get
arch = nvvm.get_arch_option(*cc)
File "C:\Users\testing\AppData\Local\Programs\Python\Python37\lib\site-packages\numba\cuda\cudadrv\nvvm.py", line 345, in get_arch_option
return 'compute%d%d' % arch
TypeError: not enough arguments for format string

Looks to me this might be a version mismatch issue
can someone help with which version we need to use in the window and Linux?
Batter if provided all prerequisite lib name and version that needed that might help.

Hi,
I tried the stitching the tool in windows. It worked. However, I went thoroughly through the paper for how the stitching works. Can you suggest some good direction on how that can be implemented efficiently like you had the binaries in windows? This project is really intriguing. Thanks.

Hi,
when I follow your work, I find that the dataset you provide donesn't incorporate the scanned images, which are used as gorund truth to train the illumination correction network.
Could you release this datavset

Thank you！

Excellent work for document rectification!

But I got some problem when using the Graphcut.exe to get the stitchedFlow.npy
terminal as this:
PS D:\abpycharm\DocProj\Stitching> .\Graphcut.exe "\test data\2_patchFlows.npy" "\test data\5.npy"
\test data\2_patchFlows.npy
\test data\5.npy
flow dir: \test data\2_patchFlows.npy

But there is no 5.npy generated. I confirmed that 2_patchFlows is regulatory size as[11,8,2,256,256]
Please help me !

Thanks for your work.
Can you share the version of numba, numpy and skimage?
my local version:
numpy:1.14.0
skimage:0.13.1
numba:0.38.0

Using @cuda.jit, a error occurred.
`numba.cuda.cudadrv.error.NvvmError: Failed to compile

libnvvm : error: -arch=compute_61 is an unsupported option
NVVM_ERROR_INVALID_OPTION`
I try to update numba to the lasest, but it requires the version of numpy laster than 1.15.0, so skimage is not compitable

Hi @xiaoyu258

Thanks for sharing the code. Could you share more details of how to use blender to render the surface flow? so that we can introduce our own dataset for training, thanks

https://colab.research.google.com/drive/1NmKyW8jceBMpfkcfrozWxdwK7JduxBrX?usp=sharing
Please tell me if there is something I'm doing wrong in the notebook

Hi, thanks for your solid work.

I test the illumination network using some geometric rectified images. The results are not very well as in paper.
Like the following examples：

We check the code is right：

I don‘t know what is the problem.
Thank you!

I can't access to that.
The message is 'This link has been removed.'
Can you restore the file? [OneDrive]

I am sorry but it is not allowed to use BitTorrent on my environment :(

I just used the sample data with ground truth npy, and use the resampling.py to unwarp the imgs. The results are not good. The background cannot be cropped.

I have trouble to download the dataset files due to big size and always timeout during download. @xiaoyu258 could you please split dataset to smaller size pieces, say 5GB?

The dewarp result in your paper is excellent. However, the links of pre-trained model and dataset are all invalid, i also try to use Blender to generate synthetic distorted document image, it is hard for me to master the software. so, could you please provided the dataset and pre-trained model again, very appreciated for your kindly reply !

请问那个光照矫正模型训练最少需要几个GPU？？

Thank you for sharing such wonderful work. Where can I get illumination network dataset?

Hello! I used your geometry_model to correct distorted images. I followed your doc and tried the test_data, everything is just right.
But then I used images from your paper, it cant get the similar result. It looks like just resized it and changed little.

input

output

I dont know how to run this graphcut,can you give me some suggestions plz

Click .exe directly come up a problem like this.

Hi @xiaoyu258, the dataset torrent provided in the README has no peers detected to download from.

Hello every one i trained the illumnation model on my dataset and when i tried to run the evaluation i am getting this error:

from torchvision import _C
ImportError: DLL load failed: Le module spécifié est introuvable.

torch version 1.1.0

@xiaoyu258
it's better that you upload the blender file, so that we can generate using our own images.
So that you wont need to upload the entire 65GB dataset. The blender file will be enough.

Hello, I'm trying to generate dataset for my own images, but got stuck on extracting the texture coordinates when given camera view coordinates. So I wonder if you are planning to open source your scripts for dataset generating?
If not, could you please give some hints on how to transform the pixel coordinates in rendered image to the UV coordinates?
Thank you !