1
$\begingroup$

Does anyone know a paper or code that does "unsupervised domain adaptation" for regression task? I saw most of the papers were benchmarked on classification tasks, not regression. I want to do something like training a model to predict a scalar value from an image (e.g. predicting image of a road to steering wheel angle for a self-driving car). One of the examples could be training on synthesis data from simulated environment (think GTA) and then trying to predict on real-world.

Here is one of the examples of unsupervised domain adaptation algorithm that also has an easy-to-access code with Keras: https://github.com/bbdamodaran/deepJDOT But it's for classification. The author said it can be used for regression but I had to change it. I changed it and it didn't work well so I don't know if it's my fault or the algo is not good for regression. I want to see papers that were benchmarked on regression so I know how well it performs on regression.

My real use case is to predict facial expression as a value from 0 to 1 like how open is the mouth. The source domain and target domain are real-world images but from different lighting.

Any suggestions are appreciated.

$\endgroup$
  • $\begingroup$ The question is asking for a data-driven approach. Which means, that a collection of example pictures is available and the neural network is able to identify a feature in the picture. The task is that the keras parameter model is transferred to a new problem. $\endgroup$ – Manuel Rodriguez Jul 6 at 9:26
  • $\begingroup$ DeepJDOT is now working well. I have contributed to the author's code and now it's working for regression as well. $\endgroup$ – off99555 Jul 20 at 4:22

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.