How to train in custom data?

[R1234A]

I have images and their respective XML files, but how to start train for our own custom dataset?

[mubastan]

First, you need to write you own data loader, similar to datasets/pascal_voc.py. Then, you should also modify datasets/factory.py accordingly. Finally, you should provide the name of your dataset as input to the trainval_net.py (--imdb and --imdbval options).

Another option is to convert your dataset into one of the standard formats, like PASCAL_VOC or MS COCO and directly use the code without any modification.

I would recommend the first option.