search

rezoo / illustration2vec

A simple deep learning library for estimating a set of tags and extracting semantic feature vectors from given illustrations.
MIT License
669 stars 112 forks source link

Create your own models #5

Open chernandez2606 opened 8 years ago

chernandez2606 commented 8 years ago

Is there a way to train your own models so that someone can add new tags?

rezoo commented 8 years ago

Sorry for replying late. I think the best way to do deal with this problem is to replace the convolutional layer before global-average pooling layer with a new one, and fine-tune it. In this case I recommend you to use caffe.

chernandez2606 commented 7 years ago

Sorry for my late reply. I'm fairly new and interested in this. Can you point me to some documentation that would lead me to the first steps for doing this? Thank you.