seanbell / intrinsic Goto Github PK

After git clone the repo from the github
I have install the dependecies as the instruction

And I also compile it successfully
But when I tried to run the script I got such error

Traceback (most recent call last):
  File "bell2014/decompose.py", line 22, in <module>
    from bell2014.solver import IntrinsicSolver
  File "/home/SENSETIME/lipenghui/mycode/intrinsic_test/bell2014/solver.py", line 8, in <module>
    from .decomposition import IntrinsicDecomposition
  File "/home/SENSETIME/lipenghui/mycode/intrinsic_test/bell2014/decomposition.py", line 3, in <module>
    from . import image_util
  File "/home/SENSETIME/lipenghui/mycode/intrinsic_test/bell2014/image_util.py", line 5, in <module>
    from skimage.filter import denoise_bilateral
ImportError: No module named filter

Here is my system configuration

Description:    Ubuntu 16.04.6 LTS
Codename:    xenial
Graphics:       GeForce GTX 1060 6GB/PCIe/SSE2
OS Type:       64-bit
Python 2.7.12 (default, Nov 12 2018, 14:36:49) 
[GCC 5.4.0 20160609] on linux2
numpy==1.16.4
scipy==1.2.1
matplotlib==2.2.4
cython==0.29.10
pillow==6.0.0
PIL==6.0.0
scikit-image==0.14.2
scikit-learn==0.20.3

So what should I do to make it work?
Have anyone encountered such problem? Any help is much appreciated.
Thx very much!

hello.just a question :
do I need these files ? and where can I get them (masks,judgements)
-m , --mask
Mask filename
-j , --judgements
Judgements file from the Intrinsic Images in the Wild
dataset
or do I just have to leave it like this as default.
thanks in advance
luc

I am running make in a conda environment on MacOS Catalina with the dependencies installed. However, many packages such as new, cerrno, cstddef and many others are not found. One of which, called Eigen, I was able to download and put into correct directory, but with so many packages not found, I am questioning whether I missed something in configuring environment. Does anyone know what's going on here? Any help is greatly appreciated.

Hi, guys. I was attempting to run the decompose.py file by command 'python decompose.py test.jpg', while I got an error like the following:
Traceback (most recent call last): File "decompose.py", line 125, in <module> r, s, decomposition = solver.solve() File "/home/jinyuxi/Documents/intrinsic/bell2014/solver.py", line 38, in solve self.initialize_intensities() File "/home/jinyuxi/Documents/intrinsic/bell2014/solver.py", line 105, in initialize_intensities samples[:, 0] *= self.params.kmeans_intensity_scale ValueError: output array is read-only
I don't know why and how to fix this issue. Is there anybody can tell me what should I do?

Hi, @seanbell

I have just make it work normally, but now i have another confusion about the memory, seems the code is very memory consuming

terminate called after throwing an instance of 'std::bad_alloc'
  what():  std::bad_alloc
Aborted (core dumped)

have you ever noticed such problem?

I have run it with png format picture

148.7 kB
321.0 kB
824.7 kB

even for such picture my 7.7 GB computer memory struggles to work

does dense crf cost the computation intrinsically?
or I didn't config it correctlly?

What should i do to make it work for my picture (1.2 MB or even 10 MB)? Downsample?

Much thx for your immediate reply.

Hello,
I have been trying to make this working with Python 3 in a Windows environment. I have fixed most of the issues with the latest eigen code, but got stuck with the DenseCRF map() function in krahenbuhl2013.pyx, where it raised an error message as below:
"...
loading reflectances...
Error loading pickled dat file in: C:\workspace\cv\intrinsic_images_in_the_wild\bell2014\energy\prob_abs_r.dat
loaded reflectances
solve...
initialization: k-means clustering with 20 centers...
clustering done (0.07672129372008385 s). intensities:
[0.02288366 0.58980664 0.99538995 0.10214907 0.48563758 0.01581242
0.24498024 0.68647472 0.33832016 0.1594008 0.07133502 0.1033111
0.0343453 0.36848925 0.02129602 0.7977585 0.19756343 0.05561196
0.08012422 0.27910016]

run: starting iteration 0/25
stage1_optimize_r: compute costs...
compute_unary_costs...
blur sigma: 43.220018509945135 pixels (image diagonal: 432.20018509945135 pixels)
compute_unary_costs: done (0.13884072930522962 s)
stage1_optimize_r: optimizing dense crf (10 iters)...
Traceback (most recent call last):

File "", line 1, in
runfile('C:/workspace/cv/intrinsic_images_in_the_wild/bell2014/decompose.py', args='C:/Temp/3663_20150501_soi.png', wdir='C:/workspace/cv/intrinsic_images_in_the_wild/bell2014')

File "C:\ProgramData\Anaconda3\lib\site-packages\spyder\utils\site\sitecustomize.py", line 705, in runfile
execfile(filename, namespace)

File "C:\ProgramData\Anaconda3\lib\site-packages\spyder\utils\site\sitecustomize.py", line 102, in execfile
exec(compile(f.read(), filename, 'exec'), namespace)

File "C:/workspace/cv/intrinsic_images_in_the_wild/bell2014/decompose.py", line 125, in
r, s, decomposition = solver.solve()

File "C:\workspace\cv\intrinsic_images_in_the_wild\bell2014\solver.py", line 47, in solve
self.stage1_optimize_r()

File "C:\workspace\cv\intrinsic_images_in_the_wild\bell2014\solver.py", line 162, in stage1_optimize_r
self.decomposition.labels_nz = densecrf.map(self.params.n_crf_iters)

File "krahenbuhl2013.pyx", line 45, in intrinsic_images_in_the_wild.bell2014.krahenbuhl2013.krahenbuhl2013.DenseCRF.map

SystemError: ..\Objects\moduleobject.c:449: bad argument to internal function"

I was not able to locate this 'moduleobject.c' file in the 'krahenbuhl2013' folder therefore couldn't go further fixing this bug. Any advice would be appreciated.

I understand that I could go for 'virtualenv' and use recommended Python 2.7 setups. But it'd be good to have this library running in Python 3 Windows env as my other code are in this env.

Thank you!

Regards,

Thank you for this great work.

I have trouble recover the input image faithfully from the decomposed shading and reflectance. I used RGB images and checked sRGB true during the decomposition process. So all output should be sRGB as well.

I perform element-wise multiplication on both shading and reflectance (both normalized to [0,1]) but the result is a lot brighter than the original.

Should I treat internet INT8 images as linear instead of sRGB during loading?

I went through the code but couldn't find anything on reconstructing the input image from the decomposed shading and reflectance, could you please remind me?

Thanks again.

Hello, thanks for your excellent work but I want to know if I can run this code faster ?
The thing is that I would like to generate a lot of shading images but it seems that it will take about 3-5 minutes to generate one shading image, which is too slow for me.

hello
when running the code on one of your example the reflectance is much brighter than the one shown on your project page.

if you look at the woman trouser for example ..
I tried to use -l (even if input is an sRGB image) but it doesn't change much

I'm running on Ubuntu 18.04 so the versions of python modules are much higher than the one required.
PIL = 5.1.0
cython = 0.26.1
numpy==1.13.3
scipy==0.19.1
scikit-image==0.14.1
scikit-learn==0.20.2

I had to change on import to have the code running
in bell2014/image_util.py :
#from skimage.filter import denoise_bilateral
from skimage.restoration import denoise_bilateral

I'm using the default params
all the tests I've done so far are giving me a very bright reflectance so I suspect some difference related to the colorspace.
thanks in advance for any advice
luc

hello, when i use "-m <mask_file>", there is an error in "decomposition.py", line 45, self.intensities[self.labels_nz].
"IndexError: index 816 is out of bounds for axis 1 with size 816"
The error only come up when i use mask，can you help me? Thank u very much!