Assertion error at LoadBatches.py for training

[matheuscass]

I'm having the following error: /LoadBatches.py", line 71, in imageSegmentationGenerator
assert( im.split('/')[-1].split(".")[0] == seg.split('/')[-1].split(".")[0] )
AssertionError

Any thoughts? =(

[giomhugo]

Am I right when I say you are on windows?
If you are on windows, you can change the / with "\\" . (To actually see what is happening, try printing im and seg)

If you're not, then sorry but I don't have any ideas.

Have a good one :)

[matheuscass]

Yes, I'm on windows. I did the following (after aplying the modification you suggested):
print(im)
print(seg)

and that was the response:

data/dataset1/images_prepped_test\0016E5_08159.png
data/dataset1/annotations_prepped_test\0016E5_08159.png
data/dataset1/images_prepped_train\0016E5_08640.png
data/dataset1/annotations_prepped_train\0016E5_08640.png

[giomhugo]

So it worked?
you forgot :
Change the assert( im.split('/')[-1].split(".")[0] == seg.split('/')[-1].split(".")[0] )
to assert( im.split('\\')[-1].split(".")[0] == seg.split('\\')[-1].split(".")[0] )

[matheuscass]

Now I'm having this error:
Traceback (most recent call last):
File "train.py", line 90, in
m.fit_generator( G , 512 , validation_data=G2 , validation_steps=200 , epochs=1 )
File "C:\Users\Matheus\Anaconda3\envs\tensorflow\lib\site-packages\keras\legacy\interfaces.py", line 91, in wrapper
return func(*args, **kwargs)
File "C:\Users\Matheus\Anaconda3\envs\tensorflow\lib\site-packages\keras\engine\training.py", line 1415, in fit_generator
initial_epoch=initial_epoch)
File "C:\Users\Matheus\Anaconda3\envs\tensorflow\lib\site-packages\keras\engine\training_generator.py", line 177, in fit_generator
generator_output = next(output_generator)
File "C:\Users\Matheus\Anaconda3\envs\tensorflow\lib\site-packages\keras\utils\data_utils.py", line 793, in get
six.reraise(value.class, value, value.traceback)
File "C:\Users\Matheus\Anaconda3\envs\tensorflow\lib\site-packages\six.py", line 693, in reraise
raise value
File "C:\Users\Matheus\Anaconda3\envs\tensorflow\lib\site-packages\keras\utils\data_utils.py", line 658, in _data_generator_task
generator_output = next(self._generator)
File "C:\Users\Matheus\Desktop\UFRGS\Mestrado\tutorial_python\APLICACAO_A_AGRICULTURA\image-segmentation-keras-master\LoadBatches.py", line 82, in imageSegmentationGenerator
im , seg = zipped.next()
AttributeError: 'itertools.cycle' object has no attribute 'next'

=(

[matheuscass]

I made the modification you mentioned, I think that assertion error is ok now, but I'm having this new error I've just mentioned above =(

[giomhugo]

Ok that error is because you are on python 3.X
modification :
itertools.cycle().next() to next(itertools.cycle())

So in your case :
zipped.next() to next(zipped)

[matheuscass]

It worked! thanks a lot for your time!!! I'm rookie on python, so it's kind of hard to solve these errors on my own... I want to use the code to apply to my own dataset, so it's likely I'll need help again...

Thanks again!

[giomhugo]

You're very welcome !
If you ever have a problem or an error just type it in google and you will surely find solutions :p

[matheuscass]

Could you answer one more question? The instructions by the author tell us to use --input_height = 320 and --input_width = 640, but these are not the exact dimensions of the images from the dataset. How those values influence on the code?

[giomhugo]

For me, they cause an error for me ! I removed these arguments
Sorry can't help you on that :(

[divamgupta]

@matheuscass there are multiple reasons that we do not use the original size of the images. If the images are very large, the the size of the CNN would also increase which means slower training time and more memory. In most cases, larger the image, you would have to use more number of layers.

[giomhugo]

@divamgupta
Let’s say my dataset are images of same dimensions (weird ratio 624*464) but the images I want to predict come from a phone (so dimensions like 1080p). What should I modify in order to get good dimensions on the prediction ?

[divamgupta]

You can first resize the image before inputting to the model and resize the output to the original dimensions