what machine learning models work best for this type of task? will it be some type of regression, or something more advanced like a neural net?
philippe-eecs
Paper is 2016 and 'deep' is in the name of the paper. They most likely use a deep CNN network, similar to that of AlexNet.
Anytime you want to deal with images in a complex space you usually use CNNs.
(CNN stands for Convolutional Neural Network)
daniswords
This is the referenced paper: https://groups.csail.mit.edu/graphics/demosaicnet/data/demosaic.pdf
One thing I found interesting is that the difficult inputs are rare, but end up with pretty bad artifacts, so the choice of training data is especially important.
what machine learning models work best for this type of task? will it be some type of regression, or something more advanced like a neural net?
Paper is 2016 and 'deep' is in the name of the paper. They most likely use a deep CNN network, similar to that of AlexNet.
Anytime you want to deal with images in a complex space you usually use CNNs.
(CNN stands for Convolutional Neural Network)
This is the referenced paper: https://groups.csail.mit.edu/graphics/demosaicnet/data/demosaic.pdf One thing I found interesting is that the difficult inputs are rare, but end up with pretty bad artifacts, so the choice of training data is especially important.