pre-processing of vis-drone dataset

[hlmhlr]

Hi,

I need some basic information regarding processing of Visdrone datatset:

1. Did you train the Faster-RCNN on visdrone for all object categories (10) and all training images (6471), or small subset of images was used based upon object categories?. If trained for few categories then how did you proceed?
2. Since, this dataset has, different size of images (140 to 350KB approx) with different resolution sizes, so how should i pre-process all theses before training. For example, to resize, to normalize, augmentation etc?
   I shall be grateful for your kind response in regard with useful steps to get the dataset ready for training, (irrespective of pascal VOC or CSV format).

Thanks,