Dataset is Noisy. Does cleaning the dataset Break Rules?

Hello organizers,
I was visualizing the dataset and found that there are many images in the training set that are black and white but present in the colored images folder.
some of such images are : 0029918fa1ddc5d7, 000c83fe629badc5, 00160895ce715368, 00241d2d17854006, 00335202286b84f0, 003d59cae7c03c39.

I have checked the mean RGB colors of such b&w images present in the colored training image folder and found that their values are the same in most of the cases. I have attached an example of such an image :

Mean RGB from colored folder :- (110.22875213623047, 110.22875213623047, 110.22875213623047, 0.0)
Mean RGB from b&w folder :- (110.22875213623047, 110.22875213623047, 110.22875213623047, 0.0))

The problem with this noise is that on training the reconstruction error for a grayscale image will be zero and that’s not useful for my model to learn.

Are the organizers aware of this issue and we have to deal with it in our way as a part of data pre-processing? And if yes, then does it break any Rule?