Closed Yuliang-Zou closed 5 years ago
I think it may be problematic to use the mean and variance of the NYU dataset on other datasets, considering that the data distribution is different. I suggest you to normalize using the mean and variance of the dataset you want to experiment on.
How can we evaluate the pre-trained models on other dataset? For example, the Make3D dataset. Can we use the mean and variance of the training set of NYU-D to normalize the predicted output?