Performance of Triplet loss network vs multiclass classification

I am training a triplet loss based classification network and a normal multiclass classification network on some image data. In my case, the triplet loss network performs worse than the multiclass network. I have tried changing layers, neurons, margin, etc. for the triplet loss network, but the multiclass network still performs better. Are there any cases where the triplet loss network can perform worse than normal multiclass classification? If no, what are the possible things I can look at to improve or change in the triplet loss network for it to work?

It is 5 class classification with some given features.

Topic classification

Category Data Science

1
answered at

For me triplet loss function (as mentioned by Neil Slater as well), is used for object recognition i.e. identify the similar object. Face-recognition is one of the use case.

This function is based upon Siamese Network, which will provide us the feature vector as an output. During recognition, we compare the feature-vector of the new image with the feature-vector of the training data. Select the one which has closer matching.

In order to improve Triplet-loss function, you can try different similarity approaches or different threshold value's. Please do share the output.

But again for me it's not comparable with object-classification use case.

About

Geeks Mental is a community that publishes articles and tutorials about Web, Android, Data Science, new techniques and Linux security.