search

PramuPerera / DeepOneClass

Deep learning based one class classification code targeting one class image classification. Tests carried out on Abnormal image detection, Novel image detection and Active Authentication reported state of the art results.
GNU Lesser General Public License v3.0
190 stars 47 forks source link

how to train on a new dataset? #6

Closed asa008 closed 5 years ago

asa008 commented 5 years ago

Sorry to bother you, but how to train on a new dataset? thank you.

PramuPerera commented 5 years ago

Hello asa008,

Training on a new dataset can be done by simply replacing the data folders with your own data. I've created two tasks 'abnormal image detection' and 'novelty detection' because their testing protocols are different. In the former, same class is divided into 'normal' and 'abnormal' subsets. Training is done on 'normal' data whereas testing is done with the objective of separating normal from abnormal data. In the latter, it's a 1 vs all problem where the objective is to separate a given class from all other classes. Most real world problems falls under the second category.