Search Results for author:
Found 9 papers, 7 papers with code
SeqTrack3D: Exploring Sequence Information for Robust 3D Point Cloud Tracking
Most existing methods perform tracking between two consecutive frames while ignoring the motion patterns of the target over a series of frames, which would cause performance degradation in the scenes with sparse points.
Motion-to-Matching: A Mixed Paradigm for 3D Single Object Tracking
3D single object tracking with LiDAR points is an important task in the computer vision field.
STTracker: Spatio-Temporal Tracker for 3D Single Object Tracking
Previous methods usually input the last two frames and use the predicted box to get the template point cloud in previous frame and the search area point cloud in the current frame respectively, then use similarity-based or motion-based methods to predict the current box.
MMF-Track: Multi-modal Multi-level Fusion for 3D Single Object Tracking
To overcome the limitations of geometry matching, we propose a Multi-modal Multi-level Fusion Tracker (MMF-Track), which exploits the image texture and geometry characteristic of point clouds to track 3D target.
Exploiting More Information in Sparse Point Cloud for 3D Single Object Tracking
Meanwhile, the encoder applies the attention on multi-scale features to compensate for the lack of information caused by the sparsity of point cloud and the single scale of features.
Real-time 3D Single Object Tracking with Transformer
PTT module in the voting stage could model the interactions among point patches, which learns context-dependent features.
3D Object Tracking with Transformer
By using cross-attention, the transformer decoder fuses features and includes more target cues into the current point cloud feature to compute the region attentions, which makes the similarity computing more efficient.
PTT: Point-Track-Transformer Module for 3D Single Object Tracking in Point Clouds
Our code is open-sourced for the robotics community at https://github. com/shanjiayao/PTT.
3D-SiamRPN: An End-to-End Learning Method for Real-Time 3D Single Object Tracking Using Raw Point Cloud
Then, to fuse the information of features in the two branches and obtain their similarity, we propose two cross correlation modules, named Pointcloud-wise and Point-wise respectively.
