We are happy to announce the first draft of our paper “A Comprehensive Study of Deep Regression”, in which we propose a sound way to evaluate (vanilla) deep regression methods. I personally think that one of the most interesting conclusions is that the data pre-processing can have much higher impact in the performance that changes in the architecture. Therefore, people out there working on deep regression, pay attention to how you pre-process your images!!!
Comments to the paper are welcome: https://arxiv.org/abs/1803.08450