Error converting images to black & white

[rael346]

I'm encountering this error when using the -bw flag

Error processing image [image path]: cannot reshape array of size 62500 into shape (3)

Not sure why this is the case, is my image size too big?

[ChenglongMa]

Hi @rael346,

Thanks for your valuable feedback!

I think it is a bug in my algorithm. Would you mind sharing this image so that I can debug it? (Sorry to keep asking for your data as I don't have a suitable dataset at hand to debug😥. I would appreciate your help!)

BTW, just confirm that -bw (--black_white) means:

Whether to convert the input to black/white image(s). If true, the app will use the black/white palette to classify the image.

It will force convert images (even if they are in color) to black and white and then sort them with black/white palette.

If you just want to set the color or black/white palette to use without converting the image, you can use option --image_type or (-t).

Thanks.

[rael346]

HI @ChenglongMa

Here is the image

Also I did try the --image-type option but when I ran it on the dataset that I'm using, it sometimes mis-classified the clothing as skin too, so I want to test if converting the images to black & white will help

[ChenglongMa]

Hi @rael346,

I have fixed the -bw error and released v1.1.2.
You can run

pip install skin-tone-classifier --upgrade

to upgrade the library.

However, I tested the algorithms on the images you provided and found that this algorithm performs poorly a few AI generated images.
The algorithm cannot effectively identify facial areas and skin areas in the image.

This is because

1. Sometimes the distribution ratio of facial features in the pictures generated by AI is incorrect;
2. Furthermore, the algorithm is trained more on portraits taken by cameras under natural light, which are often different from the pictures generated by AI.