Data split confusion

[Jiyang-Zheng]

Hi Joseph,

I have a few questions in regards to the data split. Could you please solve my confusion?

1. I can see that all data split codes are in coco_utils, however, I haven't found the file for creating 't1' imageset. Could you please instruct me where to find the file for creating 't1' imageset.

2. In 'create_tx_imageset.py', I have noticed that the image is added to the training set by if not set(classes).isdisjoint(T2_CLASS_NAMES): image_ids.append(image_details['file_name'].split('.')[0]) cls.extend(classes) This means if there is any class in the image that belongs to a known class, it will be added to the training set. I think in this case, shouldn't it be if set(classes).issubset(T2_CLASS_NAMES): image_ids.append(image_details['file_name'].split('.')[0]) cls.extend(classes) Cause we need all classes in the image to be known classes at the training phase. Or have you modified the code anywhere to remove those unknown classes from the image during the training?

Thank you and looking forward to your reply.