mixmixmix / learnmachine Goto Github PK

There is sometimes a mismatch between libraries used by different OpenCV versions on your computer.
For instance I originally had 4.1.0 and the following code for loading image works fine:

import cv2
print(cv2.__version__)
a = cv2.imread('./assets/simple_blobs/bloby_478.png')
print(type(a))

4.1.0
numpy.ndarray

However, when I created a new virtual environment with conda-forge::opencv=4.1.2 the png image won't load. The same code produces

4.1.2
NoneType

Without raising any errors.

However when running the same code in REPL we can see the warning that explains everything:
libpng warning: Application built with libpng-1.4.12 but running with 1.6.37

Different versions of opencv are built with different versions of png libraries but they are somehow forced to use the same version, or are outright broken.

That causes first cell in Task B to fail with the following error:

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-1-147208002c1c> in <module>
     16 image_size = 64  # in pixels
     17 number_of_classes = 2 #2-6 - normal blobs. 7-11 more difficult
---> 18 problem = generateBlobsData(imageDir, number_of_classes, number_of_samples, image_size, noiseSize=20)

~/repos/learnmachine/tasks/utils/practice_data.py in generateBlobsData(imageDir, noClasses, trainSamples, imSize, noiseSize, colour)
    106         #add noise
    107         if noiseSize>0:
--> 108             row,col,ch= blob_image.shape
    109             gauss = np.random.normal(0,noiseSize,(row,col,ch))
    110             gauss = gauss.reshape(row,col,ch)

AttributeError: 'NoneType' object has no attribute 'shape'

If you encounter it, try re-setting environment with different version of OpenCV.

KeyError Traceback (most recent call last)
in
16 image_size = 64 # in pixels
17 number_of_classes = 2 #2-6 - normal blobs. 7-11 more difficult
---> 18 problem = generateBlobsData(imageDir, number_of_classes, number_of_samples, image_size, noiseSize=20)

~\OneDrive - Indian Institute of Science\Side\2020-03 Machine-Learning Mix\tasks\utils\practice_data.py in generateBlobsData(imageDir, noClasses, trainSamples, imSize, noiseSize, colour)
98 print(blob_file)
99 blob_class = blob_file.split("/")[-1].split("_")[0]
--> 100 blob_class_int=intblobclass[blob_class]
101 if(noClasses <= blob_class_int):
102 continue

KeyError: 'simple'

Make a different environment for TF

print('Correct classification of Logistic Regression: {0.2%}'.format(trueRate)) # Works for 2 cell types only