Table 6.

Comparison of object pose estimation datasets.

Dataset	Years	Levels	Categories	Suitable Scenes
LM [78]	2012	Instance-Level	15	Objects are cluttered and untextured with limited viewpoints.
LM-O [79]	2014	Instance-Level	8	Objects are cluttered and more severely occluded.
Shapenet [135]	2016	Category-Level	16	Point cloud dataset of common objects in life with fine segmentation.
T-LESS [12]	2017	Instance-Level	30	Industry-related scenes with few object textures, strong symmetry, and mutual occlusion.
ITODD [13]	2017	Instance-Level	28	Industrial scenes with strong and scarce color information in the case of random projections.
Siléane [136]	2017	Instance-Level	8	Different symmetry objects.
YCB-V [20]	2018	Instance-Level	21	Daily objects with occlusion in different light situations, and applicable to the video needs of the object.
TUD-L/TYO-L [35]	2018	Instance-Level	24	Different light conditions.
NOCS [116]	2019	Category-Level	6	Category-level position of common objects, meet real and synthetic dataset requirements.
Fraunhofer [137]	2019	Instance-Level	10	Industrial large-scale dataset, including different modalities, is suitable for grasping tasks.
Objectron [138]	2021	Category-Level	9	Meeting generalizability and tracking task requirements with large-scale multiple views.