Introducing a support image layer (Lesson 3 CamVid)

Hi! I’m trying to improve the performance of segmentation by introducing Canny Edges as another input layer. Basically this finds the edges in the picture and draws them clearly. You can create it with CV2 by simply using “cv2.canny(img, x, y)”. Is it possible to use two images as inputs? One is the original image and other is the edge image to support the original one. Or would this actually confuse the algorithm?

Input images (x):

Target image (y):

Thanks for help!

You copy it to one channel of your 3-channel RGB picture or concatenate it to get a 4-channel picture.

However, it would be interesting if this really improves the performance as in deep learning you can also let the network figure the feature engineering out on its own.

If you have results I would be happy to hear about them. :slight_smile:


Ah that’s a great idea! Will try this as soon as I have the time to get back on the project, and let you know how it worked out.