Segmentation when Image Channels Are More Than 3

[ste (Stefano Giomo)]

Here you can find a starter the code to:

• load multi channel image
• adapt resent first conv-layer to accept more than 3 channels

Classifier with multiple images as input and multiple labels as output

I’ve tried to address Kaggle “Human protein atlas” competition: Where the input consist of 4 image per sample (RGB + Yellow) and the output is a multi class label, with a domain of 28 possible classes. To address this competition I’ve to implement a couple of custom function to cope with “multi channel images” datasets. 1) Multi channel image loader Inspired by fast.ai “open_image”, this function take as input a list of “Path to image” and transforms it in a single multi channel image. The…