used data

[RashoAli]

Thanks for the grate work. i have aquestion regarding the used data: in the nuscenes2d_temporal_infos_train.pkl file there are many keys ['lidar_path', 'token', 'prev', 'next', 'sweeps', 'frame_idx', 'cams', 'scene_token', 'lidar2ego_translation', 'lidar2ego_rotation', 'ego2global_translation', 'ego2global_rotation', 'timestamp', 'gt_boxes', 'gt_names', 'gt_velocity', 'num_lidar_pts', 'num_radar_pts', 'valid_flag', 'bboxes2d', 'bboxes3d_cams', 'labels2d', 'centers2d', 'depths', 'bboxes_ignore', 'visibilities']

Yet, not all the keys are used for training (as mentioned in #13), and in another instance, you stated that lidar data is not used. Therefore, lidar_path is also not used. a) which keys are used? b) The difference between gt_boxes and bboxes3d_cams? (thay both are 7 element array but there are no similarity)

thanks

[exiawsh]

'token', 'scene_token', 'lidar2ego_translation', 'lidar2ego_rotation', 'ego2global_translation', 'ego2global_rotation', 'timestamp', 'gt_boxes', 'gt_names', 'gt_velocity', 'num_lidar_pts', 'num_radar_pts', 'valid_flag', 'bboxes2d', 'labels2d', 'centers2d', 'depths',

[RashoAli]

Thanks for your fast response. i still have 2 questions: 1) if lidar is not used why are the parameters 'num_lidar_pts', 'num_radar_pts' used? 2) important parameters like the intrinsics of the cameras and image path are stord in "'cams'" so why its not used, do you get this infos from some where else?

[exiawsh]

@RashoAli

1. You don't need to download the pointcloud file in nuscenes dataset when obtain the 'num_lidar_pts', 'num_radar_pts'. The information is recorded by their api.
2. I'm sorry. It's my fault. The 'cams' info is used.

[RashoAli]

Thanks for your reply, still i have 2 question: 1) are all elements in the "cams" used? 2) what is the difference between gt_boxes and bboxes3d_cams?

[RashoAli]

Dear auther, i was checking the provided code and in the 'CustomNuScenesDataset' in 'get_data_info' you are getting just the fowloing data:

'token','lidar_path','sweeps','ego2global_rotation',ego2global_translation, lidar2ego_rotation, lidar2ego_translation, 'prev','next','scene_token','frame_idx','timestamp','bboxes2d',,'labels2d','centers2d','depths','bboxes_ignore'

and in 'cams': timestamp, sensor2lidar_rotation, sensor2lidar_translation, cam_intrinsic

dose this means that data like 'gt_boxes' is not used?

[exiawsh]

gt_boxes is used during training.

[exiawsh]

self.get_ann_info(index)

[RashoAli]

thanks for your reply agane: i belive i have just one last question regarding the coorinat system of the lables: 1 - camera_coord : 'bboxes2d', 'labels2d', 'centers2d' 2 - Lidar : 'gt_boxes', 'gt_velocity 3 - depth: distance to lidar 4 - bboxes3d_cams : 3d bbox in camera_coord

is that right so?