1
$\begingroup$

When we have an image to be used as an input to a CNN and we want to classify only part of the image, we usually feed the classifier with a crop of the image.

Lets say my image is called frame and x, y, w and h are xmin, ymin, xmax and ymax, respectively:

frame = frame[y:y + h, x:x + w] #Crop a part of the image

What does y:y or x:x mean and why do we sum them to h and w, respectively?

I've been seeing some people performing the crop in the following way:

frame = frame[y:h, x:w] #Crop a part of the image without adding to `w` and `h`

I saw the second approach being used in some places like in the following line: https://github.com/balajisrinivas/Face-Mask-Detection/blob/master/detect_mask_video.py#L51

What's the difference?

$\endgroup$
1
$\begingroup$

Lets say my image is called frame and x, y, w and h are xmin, ymin, xmax and ymax

You're confusing $w$ with $xmax$ and $h$ with $ymax$: Usually $w$ is the width of the crop whereas $xmax$ is the horizontal position of the end of the crop. Similarly $h$ is the height and $ymax$ is the vertical position of the end of the crop.

Logically since $x$ is the (horizontal) start of the crop and $w$ is the width, we can obtain $xmax$ like this: $xmax=x+w$.

Example: in a 100x100 image, let's say we want to crop a 20x20 square in the centre: $x=40, y=40, w=20, h=20, xmax=60, ymax=60$.

In the following code:

frame = frame[y:y + h, x:x + w]

the operator : is used to represent a sequence (for instance 3:7 means 3,4,5,6) so y:y + h represents the sequence from y to y+h, i.e. from $y$ to $ymax$. Same for x+w, so this line would select the part of the array corresponding to the crop.

Your second example is wrong due to the same confusion, the actual code is:

face = frame[startY:endY, startX:endX]

In this case the author is directly using the end coordinate endY (same as $ymax$) instead of calculating it as startY+h.

$\endgroup$
1
  • $\begingroup$ What an amazing explanation. Thank you very much Erwan! $\endgroup$
    – Maf
    Dec 10 '20 at 15:31

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.