Any description about the GT json files?

[JiejiangWu]

Hi, thanks for providing such a wonderful dataset. However, I have some problems.

1. About the GT json files: model_info.json, test_set.json, train_set.json Are there any descriptions about the structure of these json files? For train/test_set.json, it contains the 'annotations' part. Most contents are easy to understand, but the 'segmentation'->'counts' in each 'annotations' confuses me. More specifically, what's the meaning of data['annotations'][0]['segmentation']['counts']? Is it all x,y coords of the object mask?
2. The script 'demo_render_scene_image.py' requires the 'scene_pose_info.npy' as input. Are there any scripts to generate the 'scene_pose_info.npy', or how can I generate the 'scene_pose_info.npy' myself from the dataset?

[BwCai]

Hi, thanks for your interest. The train/test_set.json is based on the COCO format. And you can use the cocoapi to generate the object mask. I am afraid that there is no script to generate the scene_pose_info.npy. But it is easy to generate this file by yourself. You can extract all the data used in scene_pose_info.npy from the train_set.json (pose, model_id, fov part).