We aim to use RGB images to help recognizing RGB-D images.
The correlation between the two target classifiers is utilized.
The correlation between source and target domains is utilized.
A unified framework is presented to learn the classifier parameters.
Extensive experiment results show the astonishing results.