Tensorflow Multi-label Classification

General Discussion
#1

I can’t find a detailed tutorial about multi-label classification on TensorFlow site. Any resources?

#3

I recommend this video tutorial at
https://medium.com/deep-learning-with-keras/which-activation-loss-functions-in-multi-class-clasification-4cd599e4e61f

#5

Thanks @Takashi_Futada, But I’m looking for Multi-label Classification tutorial not Multi-Class

#6

He mentioned multi-labels as well as multi-classes in this article. There is no entire codes but some snippet with multi-hot encoding.
https://medium.com/deep-learning-with-keras/how-to-solve-classification-problems-in-deep-learning-with-tensorflow-keras-6e39c5b09501

#8

AFAIK, The closest thing we have on tensorflow.org right now is the segmentation tutorial, just because it, effectively, runs a separate classifier per pixel.

THere’s nothing to stop you from using an extra dimension in your outputs and labels to run a bunch of classifiers in parallel.

But also note that if your model returns a distionary of tensors, keras’s model.fit will also accept a dictionary of losses and loss_weights (it optimizes the weighted sum of the losses). So if you need 10 binary classifiers, 3 different-N N-way classifiers, and a M-element regression you can do that from a single model.fit.

1 Like