1

I'm trying to convert a dataset of dimensions (32, 32, 3, 10000) dimension dataset to a grayscale dataset, where I would have (32, 32, 1, 10000) dimensions, but I need to have that 1 channel because I will input this to a Neural Network. I tried using numpy.average, but the shape becomes (32, 32, 10000) which the TensorFlow unit is not taking as an input. I even tried to manually average it, but it had the same result. Could you guys help me with this?

ASChamp
  • 13
  • 3
  • Does this answer your question? [How to convert RGB images to grayscale, expand dimensions of that grayscale image to use in InceptionV3?](https://stackoverflow.com/questions/65179822/how-to-convert-rgb-images-to-grayscale-expand-dimensions-of-that-grayscale-imag) – Dhana D. Aug 19 '21 at 11:01

2 Answers2

0

Try (0.299 * Red) + (0.587 * Green) + (0.114 ∙ Blue) instead of averaging.

Yasin Arabi
  • 90
  • 9
0

It is possible to add that extra dimension you need after getting (32, 32, 1000) as shape.

You could try np.expand_dims with the axis parameter to define where you want that extra "1" to appear.

Bedir Yilmaz
  • 3,823
  • 5
  • 34
  • 54