@beniz Thanks for the answer.
Another question when I want to make prediction for an image model. I put images in a folder however there might some over images files that are not images), I seed that it fails because it can not resize an object that is not an image. However when I train and make the lmdb those files are ignored, is there any way to do the same thing for prediction (ignore the files that are not images) directly with DD. Or do you plan on implementing it?