fatheral / matlab_imresize Goto Github PK

Thank you very much. :)

My input image shape is [32,3,64,64], but the shape of result is [64,6,64,64] while it should be [32,3,128,128]
Did my modification cause it?
When I run the original code, a mistake happened: ValueError: operands could not be broadcast together with shapes (64,4,1,1) (64,4,3,64,64)
The mistake happens in the sentence: outimg = np.sum(weights*((inimg[indices].squeeze(axis=1)).astype(np.float64)), axis=1)

def imresizevec(inimg, weights, indices, dim):
wshape = weights.shape
if dim == 0:
weights = weights.reshape((wshape[0], wshape[2], 1, 1))
print(weights.shape)
outimg = np.sum(weights*((inimg[indices].squeeze(axis=1)).astype(np.float64)), axis=1)
print(outimg.shape)

So I change the "weights = weights.reshape((wshape[0], wshape[2], 1, 1))" to "weights = weights.reshape((wshape[0], wshape[2], 1, 1,1))"
Could you help me? Thanks a lot.

According to https://stackoverflow.com/a/58363402/6065356 it's not enough to change the parameter in OpenCV since they use round in contrast to MATLAB which uses ceil to deduce the dimensions of the image. Anyways nice implementation which matches the MATLAB imresize.

I was interested in using your library as it mimics Matlab's imresize perfectly, however I noticed that there was no license attached to the project/repo.

Github's default policy for this is (taken from https://help.github.com/en/github/creating-cloning-and-archiving-repositories/licensing-a-repository):

without a license, the default copyright laws apply, meaning that you retain all rights to your source code and no one may reproduce, distribute, or create derivative works from your work

Have you considered adding a license so that others (including myself) can use your source code in our own projects?

Thanks!

since sometimes it is deprecated we can change the main file

import numpy as np
from scipy import ndimage
import matplotlib.pyplot as plt
from imresize import *


img_uint8 = ndimage.imread("test\\lena_512x512.png")
plt.figure()
plt.imshow(img_uint8)
new_size = (123, 234)
new_img_uint8 = imresize(img_uint8, output_shape=new_size)

plt.figure()
plt.imshow(new_img_uint8)

plt.show()

I am using your code.
But i can not get the same psnr with matlab's.
This is my code (https://github.com/spnova12/get_set5_psnr)
If you do not mind, please check my code.
I also put the matlab code I used on github together.

I read your article posted at https://www.reddit.com/r/MachineLearning/comments/6vdo51/p_matlab_bicubic_imresize_implemented_in_python/.

I totally agree with you that using matlab enforces off-line dataset building. (bad!!!)
You commented that "Note: it is not very optimized in terms of speed, and probably the best solution is to re-compile OpenCV (there single constant for cv::resize should be changed). But it is not always appropriate for all.", I have two questions with this one.

1. What single constant should I change to use OpenCV bicubic similar to matlab version?
2. Does OpenCV bicubic show similar performance (e.g. PSNR) to matlab bicubic?

Thank you,
Hyunho.