Hi,
I am doing a research project related to Face Recognition and using LFW dataset.
If I import LFW as colour = True using some additional parameters (see attached word document), the CNN architecture gets built smoothly.
However, if I import LFW as colour = False, using the same additional parameters, I am getting an error. Apparently, it's to do with Conv2D layer not able to read input in grayscale.
I have added the Validation message as well. Can somebody please help me either fix it or explain why this is happening.
Many thanks
Emanuel
Looking at the docx uploaded by you, This is a simple dimensionality error. you need to get dimensions of your input right.
you have specified input dimensions as
input = input(shape=X.shape[1:])
X.shape[1:] is (154,154,3). so you configured your model that this would be the input to this model. However when you input gray image, dimensions of the input changes. probably (154,154,1) hence the error.
The model expects that the image input layer has the dimension of [154, 154, 3] (RGB image).
So, you need either to tweak this to [154, 154, 1] as suggested above or if you don't want to change your constructed model, fill the RGB three tuples with duplicates of the grayscale image.
It is mentioned in the ValueError, the excepted input is to have 4 dimensions. The color image that you passing has 4 dimensions, but with the grayscale image, the dimension reduces to 3. Therefore, it is needed to expand the dimension of the array at axis = -1. The same has been pointed in other answers too.
- Badges
- Science topic
- Similar topics
- Design
- Architecture