2021 IEEE Virtual Reality and 3D User Interfaces (VR) 2021
DOI: 10.1109/vr50410.2021.00042
|View full text |Cite
|
Sign up to set email alerts
|

SuperPlane: 3D Plane Detection and Description from a Single Image

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
2
1
1

Citation Types

0
3
0

Year Published

2022
2022
2024
2024

Publication Types

Select...
2
2

Relationship

2
2

Authors

Journals

citations
Cited by 4 publications
(4 citation statements)
references
References 33 publications
0
3
0
Order By: Relevance
“…Specifically, when creating AR work that involve the recognition of multiple AR marker and require specific logical controls based on these AR markers, image-based recognition is the preferred choice. However, in scenarios where an AR work is desired without the need for complex marker-based logic, plane detection is more suitable because it is still a challenge to extract robust features in weakly texture scenes [28].…”
Section: ) Model Registration Methodsmentioning
confidence: 99%
“…Specifically, when creating AR work that involve the recognition of multiple AR marker and require specific logical controls based on these AR markers, image-based recognition is the preferred choice. However, in scenarios where an AR work is desired without the need for complex marker-based logic, plane detection is more suitable because it is still a challenge to extract robust features in weakly texture scenes [28].…”
Section: ) Model Registration Methodsmentioning
confidence: 99%
“…In instance tracker, the instance embedding is trained to track the elements between frames. Existing methods use triplet loss [8] or contrast loss [5] to pull the same instance as close as possible in feature space and push different instances as far as possible. These methods establish instance correspondence by identifying nearest neighbour matches in feature space, which is non-differentiable and unable during the training process.…”
Section: Instance-wise Differentiable Matching Layermentioning
confidence: 99%
“…To solve the problem of training convergence issue that usually exists for the instance-based tracker [7,8], we propose a differentiable matching layer to obtain the correspondence between different instances and then exploit the cross-entropy loss to train the network. In pixel tracker, rather than sparse objects' centre in CenterTrack [6], we use the dense optical flow estimation network RAFT [9] to obtain the dense pixel motion of the current frame and add the original coordinate of pixels to obtain the mask of each instance in the next frame.…”
Section: Introductionmentioning
confidence: 99%
“…S IMULTANEOUS localization and mapping (SLAM) is one of the most fundamental tasks in the field of computer vision and robotics, with applications ranging from augmented reality (AR), virtual reality (VR) to autonomous driving. In AR applications, SLAM is often used to provide accurate localization to facilitate users to place virtual objects [1], while the dense reconstruction is increasingly needed for better interaction with the environment. Monocular dense SLAM [2], [3] has received much attention due to the simplicity of monocular video acquisition, yet it is a much more difficult task compared to RGB-D SLAM [4]- [8].…”
Section: Introductionmentioning
confidence: 99%