- Published on
Paper review: CenterTrack - CenterPoint - CenterPoint++
- Viet Anh
This is my note for 3 papers: CenterTrack - 2D object tracking, CenterPoint - 3D object detection and tracking, and CenterPoint++ - submission to the Waymo Real-time 3D Detection Challenge.
CenterTrack applies a detection model to a pair of images and detections from the prior frame. Given this minimal input, CenterTrack localizes objects and predicts their associations with the previous frame paralelly with detection phase. By that way, the cost for tracking is very cheap. This idea is also applied in CenterPoint and CenterPoint++.
Besides video data, CenterTrack can also be trained on static images. When training on video data, they add random noise into video frame and detection results. To train this network on static images, they generate previous frame by applying random scaling and translation on current frame.