6D object pose estimation
SHREC 2020 Track
6D pose estimation is crucial for augmented reality, virtual reality, robotic grasping and manipulation and autonomous navigation. However, the problem is challenging due to the variety of objects in the real world. They have varying 3D shape and the appearances of captured images from them are affected by sensor noise, changing lighting conditions and occlusions between objects. With the emergence of cheap RGB-D sensors, the precision of 6D object pose estimation is improved for both rich and low textures objects. Nonetheless, existing methods have difficulty to meet the requirement of accurate 6D pose estimation and fast inference simultaneously.
In this SHREC track, we propose a task of 6D pose estimate from RGB-D images in real time. We provide 3D datasets which contain RGB-D images, point clouds of eight objects and ground truth 6D poses. We hope that this will enable researchers to try out different methods.
To provide participants with as accurate ground truth information as possible, we have created a physically accurate simulator that is able to generate photot-realistic color-and-depth image pairs.
The dataset has eight rich and low textures objects. Each object has color-and-depth image pairs which have a resolution of 1280*720. The total number of images for each object is 500 (.png). Based on image-based rendering, we generate 400 photo-realistic synthesized color-and-depth image pairs for training and use the remaining 100 images for testing. The ground truth pose(from global to camera coordinates) is estimated by structure from motion(SFM). The Figure 1 shows eight objects of the dataset.Download the datasets
To participate in the track, please send us an email. In it, please confirm your interest in participation and if applicable, please also mention your affiliation and co-authors.
From participants, no later than the deadline mentioned in the schedule, we expect results submitted along with a one-page description of the method used to generate them. Results should be presented as a .txt file containing three parts: image names, their estimated 6D poses and estimation speed(ms) per pose.
We use two metrics for evaluation. Given the ground truth rotation R and translation T and estimated rotation R_e and translation T_e, the average distance(ADD) metric computes the mean distances between each 3D model points transformed by [R_e|T_e] and [R|T].
The average closest point distance (ADD-S) is an ambiguity-invariant pose error metric which takes care of both symmetric and non-symmetric objects into an overall evaluation.
- Honglin Yuan 1
- Remco C. Veltkamp 1
The registration and submission deadlines are in AoE (Anywhere on Earth) timezone.