How to convert 2D array one hot encoding to class name label?

Question

I have just trained a CNN model for rock-paper-scissors image classification with Tensorflow. If I try to predict a class of an image, it gives output like [[1. 0. 0.]]. How to convert it to a class name label? I use ImageDataGenerator() method with class_mode="categorical" argument.

Answer 1

You need to decode the output. A basic one can be as follows:

import tensorflow as tf

BATCH_SIZE = 3
NUM_CLASSES = 3

one_hot_encoded = tf.constant([[0, 1, 0],
                               [1, 0, 0],
                               [0, 0, 1]])

# Compute the argmax across the columns.
decoded = tf.argmax(one_hot_encoded, axis=1)