CODEWITHSUNDEEP
tensorflowkerasdeep-learningconv-neural-networkmultilabel-classification
JAGANNADAN MANASA
JAGANNADAN MANASA

Reputation: 9

How to implement Multi label classification with train, validation and test images with Densenet/ResNet

I am working on Multi-Label Image classification using a dataset having 14,720 training images. I need to implement it using ResNet/DenseNet. I trying to work on it. Could you please suggest me a reference to move forward.

Upvotes: 1

Views: 3488

Answers (1)

Timbus Calin
Timbus Calin

Reputation: 15033

For creating a multi-label classification problem, you have to bear in mind two different crucial aspects:

  1. The activation function to be used is sigmoid, not softmax, like in the multi-class classification problem.
  2. A correct label would be of the form [1,0,1,0,0]; practically, since we have a multi-label, we do not have the mutual exclusiveness case (in fact, that is the explanation, a more mathematically complex one for choosing sigmoid and not softmax, with regard to Bernoulli Distributions).

You can have a look here on how to create a multi-label classification problem in Keras:

https://www.pyimagesearch.com/2018/05/07/multi-label-classification-with-keras/

Upvotes: 2

Related Questions