Different keypoints for different objets in custom dataset

[ErikDerGute]

Hello,

Iam new to mmpose. After some time reading the docs I think I roughly figured out how to use the toolbox. However, I want to train a rsn model from scratch using a custom dataset. I generated my synthetic data using my custom written "Keypoint Annotator" for blenderproc. As I use blenderproc my annotations are in the coco format.

1. Question
   How do i configure the dataset_info file? My images contain different objects with different keypoints so my annotations look something like

..."categories": [{"id": 1, "supercategory": "pinzette", "name": "pinzette_1", "keypoint": ["kp_1_pinzette_1", "kp_2_pinzette", "kp_3_pinzette", "kp_4_pinzette_1", "kp_5_pinzette_1", "kp_6_pinzette_1", "kp_7_pinzette_1"]}, {"id": 2, "supercategory": "nadelhalter", "name": "nadelhalter", "keypoint": ["kp_1_nadelhalter", "kp_2_nadelhalter", "kp_3_nadelhalter", "kp_4_nadelhalter", "kp_5_nadelhalter", "kp_6_nadelhalter", "kp_7_nadehalter"]}],...

How do i specify different keypoints for different objects (category ids) in the dataset_info file?

2. Question
   Is it possible to skip the dataset_info file, since the annotations file basically contains all the informations of dataset_info? Like reading directly from the annotations file?

I also add an example (annotations) file to this post.
coco_annotations.json

Maybe someone could help me, thanks!