Tracking Objects as Points

By Xingyi Zhou, Vladlen Koltun, Philipp Krahenbuhl. European Conference on Computer Vision, 2020

1,379 citations recorded by the source corpus.

Tracking has traditionally been the art of following interest points through space and time. This changed with the rise of powerful deep networks. Nowadays, tracking is dominated by pipelines that perform object detection followed by temporal association, also known as tracking-by-detection. In this paper, we present a simultaneous detection and tracking algorithm that is simpler, faster, and more accurate than the state of the art. Our tracker, 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. That's it. CenterTrack is simple, online (no peeking into the future), and real-time. It achieves 67.3% MOTA on the MOT17 challenge at 22 FPS and 89.4% MOTA on the KITTI tracking benchmark at 15 FPS, setting a new state of the art on both datasets. CenterTrack is easily extended to monocular 3D tracking by regressing additional 3D attributes. Using monocular video input, it achieves 28.3% AMOTA@0.2 on the newly released nuScenes 3D tracking benchmark, substantially outperforming the monocular baseline on this benchmark while running at 28 FPS.

Subfields: Aerospace Engineering, Computer Vision and Pattern Recognition, Geology

View Paper: https://www.semanticscholar.org/paper/d58f07bab3ff01a002d139062ad83d5818c23864

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Aerospace Engineering

In Computer Vision and Pattern Recognition

In Geology

Citations in the corpus

TrackFormer: Multi-Object Tracking with Transformers (2021, methodology): “However, TrackFormer and point-based methods such as CenterTrack [22] require a procedure for filtering track initializations by public detections in a comparable manner.”
FairMOT: On the Fairness of Detection and Re-identification in Multiple Object Tracking (2020, methodology): “CenterTrack (Zhou et al. 2020) is also related to our work since it also uses center-based object detection framework.”
Neural Scene Graphs for Dynamic Scenes (2020, methodology): “Specifically, we compare renderings from scene graphs learned from KITTI [11] annotations and the following two tracking methods: we combine PointRCNN [30], a lidar point cloud object detection method, with the 3D tracker AB3DMOT [41], and we evaluate on labels from a camera-only tracker CenterTrack”
Track to Detect and Segment: An Online Multi-Object Tracker (2021, methodology): “Implementation Details: We adopt the same experimental settings as CenterTrack [63], such as backbone, image sizes, pretraining, score thresholds, etc.”
SG-Net: Spatial Granularity Network for One-Stage Video Instance Segmentation (2021, methodology): “The work similar to us is CenterTrack [46] using box centers.”
DanceTrack: Multi-Object Tracking in Uniform Appearance and Diverse Motion (2021, methodology): “Methods MOT17 DanceTrack (Proposed Dataset) HOTA DetA AssA MOTA IDF1 HOTA DetA AssA MOTA IDF1 CenterTrack [40] 52.”
Observation-Centric SORT: Rethinking SORT for Robust Multi-Object Tracking (2022, methodology): “Adding other cues such as center distance [80] or appearance similarity has been demonstrated [70] efficient to solve this.”
Quasi-Dense Similarity Learning for Multiple Object Tracking (2020, methodology): “2 759 570 22284 160491 5529 CenterTrack∗ [54] 67.”
Joint Object Detection and Multi-Object Tracking with Graph Neural Networks (2021, background): “[22]–[25] proposed to unify object detector with a model-free single-object tracker, where the tracker directly regresses the location of each object detected in the previous frame to the current frame.”
Center-based 3D Object Detection and Tracking (2020, methodology): “We show that center-based representations [61, 62] are an ideal fit for 3D application.”
PointAugmenting: Cross-Modal Augmentation for 3D Object Detection (2021, methodology): “To be specific, we take the output activation from DLA34 [30] of CenterNet [33, 32] as image features, where the channel number of the feature map is 64 and its scale factor is 4.”
Rethinking the Competition Between Detection and ReID in Multiobject Tracking (2020, methodology): “The third one is the other joint detection and tracking method without using ReID for data association, including CTrackerV1 [22], TubeTK [23], Tracktor++ [24] and CenterTrack [25].”
StrongSORT: Make DeepSORT Great Again (2022, methodology): “Extensive experiments demonstrate that they can create notable improvements in StrongSORT and other state-of-the-art trackers, e.g., Center-Track [66], TransTrack [45] and FairMOT [64], with running speeds of 1.7 ms and 7.1 ms per image, respectively, on MOT17.”
VoxelNeXt: Fully Sparse VoxelNet for 3D Object Detection and Tracking (2023, background): “CenterPoint [57] predicts the velocities to associate object centers through multiple frames, following CenterTrack [60].”
BoT-SORT: Robust Associations Multi-Pedestrian Tracking (2022, methodology): “For ablation studies, we follow [62, 58], by using the ﬁrst half of each video in the training set of MOT17 for training and the last half for validation.”