Awesome
Awesome Multiple object Tracking:
A curated list of multi-object-tracking and related area resources. It only contains online methods. 中文版更为详细,具体查看仓库根目录下的README-zh.md文件。
<a id="markdown-contents" name="contents"></a>
Contents
<!-- TOC --> <!-- /TOC --><a id="markdown-review-papers" name="review-papers"></a>
Review papers
Multiple Object Tracking: A Literature Review [paper]
Deep Learning in Video Multi-Object Tracking: A Survey [paper]
Tracking the Trackers: An Analysis of the State of the Art in Multiple Object Tracking [paper]
Machine Learning Methods for Data Association in Multi-Object Tracking [paper]
MOTChallenge: A Benchmark for Single-camera Multiple Target Tracking [paper] new paper for new MOT researcher
Multiple Object Tracking in Deep Learning Approaches:A Survey [paper]
<a id="markdown-algorithm-papers" name="algorithm-papers"></a>
Algorithm papers
<a id="markdown-2024" name="2024"></a>
2024
PASTA: Is Multiple Object Tracking a Matter of Specialization? [paper] [code] NeurIPS 2024
Hybrid-SORT: Hybrid-SORT: Weak Cues Matter for Online Multi-Object Tracking [paper] [code] AAAI 2024
UCMCTrack: UCMCTrack: Multi-Object Tracking with Uniform Camera Motion Compensation [paper] AAAI 2024
<a id="markdown-2023" name="2023"></a>
2023
ColTrack: Collaborative Tracking Learning for Frame-Rate-Insensitive Multi-Object Tracking [paper] ICCV2023
MeMOTR: MeMOTR: Long-Term Memory-Augmented Transformer for Multi-Object Tracking [code] [paper] ICCV2023
TrackFlow: TrackFlow: Multi-Object Tracking with Normalizing Flows [paper] ICCV2023
MotionTrack: Learning Robust Short-term and Long-term Motions for Multi-Object Tracking [paper] CVPR2023
C-BIoU: Hard to Track Objects with Irregular Motions and Similar Appearances? Make It Easier by Buffering the Matching Space [paper] WACV 2023
GHOST: Simple Cues Lead to a Strong Multi-Object Tracker [code] [paper] CVPR2023
<a id="markdown-2022" name="2022"></a>
2022
MTracker: Robust Multi-Object Tracking by Marginal Inference [code] [paper] ECCV 2022
Unicorn: Towards Grand Unification of Object Tracking [code] [paper] ECCV2022 Oral
P3AFormer: Tracking Objects as Pixel-wise Distributions [code] [paper] ECCV2022 oral
BoT-SORT: BoT-SORT: Robust Associations Multi-Pedestrian Tracking [code] [paper]
SGT: Detection Recovery in Online Multi-Object Tracking with Sparse Graph Tracker [code] [paper]
LPT: Learning of Global Objective for Network Flow in Multi-Object Tracking [code] [paper] CVPR 2022
MeMOT: MeMOT: Multi-Object Tracking with Memory [paper] CVPR2022 oral
UTT: Unified Transformer Tracker for Object Tracking [code] [paper] CVPR2022
OC-SORT: Observation-Centric SORT: Rethinking SORT for Robust Multi-Object Tracking [code] [paper]
GTR: Global Tracking Transformers [code] [paper] CVPR 2022
StrongSORT: StrongSORT: Make DeepSORT Great Again [code] [paper]
MAA: Modelling Ambiguous Assignments for Multi-Person Tracking in Crowds [[code]] [paper]
CrowdTrack: On the Performance of Crowd-Specific Detectors in Multi-Pedestrian Tracking [code] [paper]
<a id="markdown-2021" name="2021"></a>
2021
ByteTrack: ByteTrack: Multi-Object Tracking by Associating Every Detection Box [code] [paper] new SOTA
PCAN Prototypical Cross-Attention Networks for Multiple Object Tracking and Segmentation [code] [paper] NeurIPS2021
PermaTrack: Learning to Track with Object Permanence [code] [paper] ICCV2021
TMOH: Improving Multiple Pedestrian Tracking by Track Management and Occlusion Handling [code] [paper] CVPR2021
SOTMOT: Improving Multiple Object Tracking with Single Object Tracking [code] [paper] CVPR2021
LPC_MOT: Learning a Proposal Classifier for Multiple Object Tracking [code] [paper] CVPR2021
MTP: Discriminative Appearance Modeling with Multi-track Pooling for Real-time Multi-object Tracking [code] [paper] CVPR2021
TADAM: Online Multiple Object Tracking with Cross-Task Synergy [code] [paper] CVPR2021
RelationTrack: RelationTrack: Relation-aware Multiple Object Tracking with Decoupled Representation [[code]] [paper]
MOTR: MOTR: End-to-End Multiple-Object Tracking with TRansformer [code] [paper]
OMC: One More Check: Making "Fake Background" Be Tracked Again [code] [paper] AAAI2022
QDTrack: Quasi-Dense Similarity Learning for Multiple Object Tracking [code] [paper] CVPR2021
SiamMOT: SiamMOT: Siamese Multi-Object Tracking [code] [paper] CVPR2021
GMTracker: Learnable Graph Matching: Incorporating Graph Partitioning with Deep Feature Learning for Multiple Object Tracking [code] [paper] CVPR2021
ArTIST: Probabilistic Tracklet Scoring and Inpainting for Multiple Object Tracking [paper] CVPR2021
CorrTracker/TLR: Multiple Object Tracking with Correlation Learning [code] [paper] CVPR2021
TransMOT:Spatial-Temporal Graph Transformer for Multiple Object Tracking [code] [paper]
TransCenter: TransCenter: Transformers with Dense Queries for Multiple-Object Tracking [code] [paper]
GCNet: Global Correlation Network: End-to-End Joint Multi-Object Detection and Tracking [code] [paper]
TraDes: Track to Detect and Segment: An Online Multi-Object Tracker [code] [paper] CVPR2021
DEFT: DEFT: Detection Embeddings for Tracking [code] [paper]
TrackMPNN: TrackMPNN: A Message Passing Graph Neural Architecture for Multi-Object Tracking [code] [paper]
TrackFormer: TrackFormer: Multi-Object Tracking with Transformers [code] [paper]
<a id="markdown-2020" name="2020"></a>
2020
ReMOTS: ReMOTS: Self-Supervised Refining Multi-Object Tracking and Segmentation (1st-place solution for CVPR 2020 MOTS Challenge)[paper]
TransTrack: TransTrack: Multiple-Object Tracking with Transformer [code] [paper]
TPAGT: Tracklets Predicting Based Adaptive Graph Tracking [paper] original FGAGT
MLT: Multiplex Labeling Graph for Near-Online Tracking in Crowded Scenes [paper]
GSDT: Joint Object Detection and Multi-Object Tracking with Graph Neural Networks [code] [paper]
SMOT: SMOT: Single-Shot Multi Object Tracking [paper]
CSTrack: Rethinking the competition between detection and ReID in Multi-Object Tracking [code][paper]
MAT: MAT: Motion-Aware Multi-Object Tracking [paper]
UnsupTrack: Simple Unsupervised Multi-Object Tracking [paper]
FairMOT: FairMOT: On the Fairness of Detection and Re-Identification in Multiple Object Tracking [code][paper] a new version of FairMOT, compared with new method like CTracker
DMM-Net: Simultaneous Detection and Tracking with Motion Modelling for Multiple Object Tracking [code][paper]
SoDA: SoDA: Multi-Object Tracking with Soft Data Association [[code]][paper]
CTracker: Chained-Tracker: Chaining Paired Attentive Regression Results for End-to-End Joint Multiple-Object Detection and Tracking [code][paper]
MPNTracker: Learning a Neural Solver for Multiple Object Tracking [code][paper]
UMA: A Unified Object Motion and Affinity Model for Online Multi-Object Tracking [code][paper]
RetinaTrack: Online Single Stage Joint Detection and Tracking [[code]][paper]
FairMOT: A Simple Baseline for Multi-Object Tracking [code][paper]
TubeTK: TubeTK: Adopting Tubes to Track Multi-Object in a One-Step Training Model [code][paper]
CenterTrack: Tracking Objects as Points [code][paper]
Lif_T: Lifted Disjoint Paths with Application in Multiple Object Tracking [code][paper]
PointTrack: Segment as points for efficient online multi-object tracking and segmentation [code][paper]
PointTrack++: PointTrack++ for Effective Online Multi-Object Tracking and Segmentation [code][paper]
FFT: Multiple Object Tracking by Flowing and Fusing [paper]
MIFT: Refinements in Motion and Appearance for Online Multi-Object Tracking [code][paper]
EDA_GNN: Graph Neural Based End-to-end Data Association Framework for Online Multiple-Object Tracking [code][paper]
GNMOT: Graph Networks for Multiple Object Tracking [code][paper]
<a id="markdown-2019" name="2019"></a>
2019
Tracktor/Tracktor++: Tracking without bells and whistles [code][paper]
DeepMOT: How To Train Your Deep Multi-Object Tracker [code][paper]
JDE: Towards Real-Time Multi-Object Tracking [code][paper]
MOTS: MOTS: Multi-Object Tracking and Segmentation[paper]
FANTrack: FANTrack: 3D Multi-Object Tracking with Feature Association Network [code][paper]
FAMNet: Joint Learning of Feature, Affinity and Multi-dimensional Assignment for Online Multiple Object Tracking[paper]
<a id="markdown-2018" name="2018"></a>
2018
DeepCC: Features for Multi-Target Multi-Camera Tracking and Re-Identification [paper]
SADF: Online Multi-Object Tracking with Historical Appearance Matching and Scene Adaptive Detection Filtering [paper]
DAN: Deep Affinity Network for Multiple Object Tracking [code][paper]
DMAN: Online Multi-Object Tracking with Dual Matching Attention Networks [code][paper]
BeyondPixels: Beyond Pixels: Leveraging Geometry and Shape Cues for Online Multi-Object Tracking [code][paper]
MOTDT: Real-time Multiple People Tracking with Deeply Learned Candidate Selection and Person Re-Identification [code][paper]
DetTA: Detection-Tracking for Efficient Person Analysis: The DetTA Pipeline [code][paper]
V-IOU: Extending IOU Based Multi-Object Tracking by Visual Information [code][paper]
<a id="markdown-2017" name="2017"></a>
2017
DeepSORT: Simple Online and Realtime Tracking with a Deep Association Metric [code][paper]
NMGC-MOT: Non-Markovian Globally Consistent Multi-Object Tracking [code][paper]
IOUTracker: High-Speed tracking-by-detection without using image information [code][paper]
RNN_LSTM: Online Multi-Target Tracking Using Recurrent Neural Networks [code][paper]
D2T: Detect to Track and Track to Detect [code][paper]
RCMSS: Online multi-object tracking via robust collaborative model and sample selection [paper]
CIWT: Combined image-and world-space tracking in traffic scenes [code][paper]
<a id="markdown-2016" name="2016"></a>
2016
SORT: Simple online and realtime tracking [code][paper]
POI: POI: Multiple Object Tracking with High Performance Detection and Appearance Feature [code]
<a id="markdown-datasets" name="datasets"></a>
Datasets
<a id="markdown-surveillance-scenarios" name="surveillance-scenarios"></a>
Surveillance Scenarios
PETS 2009 Benchmark Data [url]<br> MOT Challenge [url]<br> UA-DETRAC [url]<br> WILDTRACK [url]<br> NVIDIA AI CITY Challenge [url]<br> VisDrone [url]<br> JTA Dataset [url]<br> Path Track [url]<br> TAO [url]<br> GMOT40 [url]<br> TAO-OW [url]<br>
<a id="markdown-driving-scenarios" name="driving-scenarios"></a>
Driving Scenarios
KITTI-Tracking [url]<br> APOLLOSCAPE [url]<br> APOLLO MOTS [url]<br> Omni-MOT [url]<br> BDD100K [url]<br> Waymo [url]<br>
<a id="markdown-sport-scenarios" name="sport-scenarios"></a>
Sport Scenarios
SoccerNet Minimap Tracking [url]<br> SoccerNet Tracking [url]<br>
<a id="markdown-metrics" name="metrics"></a>
Metrics
metric | formula |
---|---|
accuracy | $ Accuracy = {{TP + TN} \over {TP + TN + FP + FN}} $ |
recall | $ Recall = {TP \over {TP + FN}} = TPR$ |
precision | $ Precision = {TP \over {TP + FP}} $ |
MA | $ MA = {FN \over {TP + FN}} $ |
FA | $ FA = {FP \over {TP + FP}} $ |
MOTA | $MOTA = 1 - {\sum_t(FN + FP + IDs)\over \sum_t gt}$ |
MOTP | $ MOTP = {\sum_{t,i}d_t^i \over \sum_tc_t }$ |
IDP | $ IDP = {IDTP \over {IDTP + IDFP}} $ |
IDR | $ IDR = {IDTP \over {IDTP + IDFN}} $ |
IDF1 | $ IDF1 = {2 \over {{1 \over IDP} + {1 \over IDR}}} = {2IDTP \over {2IDTP + IDFP + IDFN}} $ |
<a id="markdown-benchmark-results" name="benchmark-results"></a>
Benchmark Results
<a id="markdown-mot16" name="mot16"></a>
MOT16
Rank | Model | MOTA | Paper | Year |
---|---|---|---|---|
1 | FairMOT | 68.7 | A Simple Baseline for Multi-Object Tracking | 2020 |
2 | JDE | 64.4 | Towards Real-Time Multi-Object Tracking | 2019 |
3 | Lif_T | 61.3 | Lifted Disjoint Paths with Application in Multiple Object Tracking | 2020 |
4 | MPNTrack | 58.6 | Learning a Neural Solver for Multiple Object Tracking | 2020 |
5 | DeepMOT-Tracktor | 54.8 | How To Train Your Deep Multi-Object Tracker | 2019 |
6 | TNT | 49.2 | Exploit the Connectivity: Multi-Object Tracking with TrackletNet | 2018 |
7 | GCRA | 48.2 | Trajectory Factory: Tracklet Cleaving and Re-connection by Deep Siamese Bi-GRU for Multiple Object Tracking | 2018 |
8 | FWT | 47.8 | Fusion of Head and Full-Body Detectors for Multi-Object Tracking | 2017 |
9 | MOTDT | 47.6 | Real-time Multiple People Tracking with Deeply Learned Candidate Selection and Person Re-Identification | 2018 |
10 | NOMT | 46.4 | Near-Online Multi-target Tracking with Aggregated Local Flow Descriptor | 2015 |
11 | DMMOT | 46.1 | Online Multi-Object Tracking with Dual Matching Attention Networks | 2019 |
<a id="markdown-mot17" name="mot17"></a>
MOT17
Rank | Model | MOTA | Paper | Year |
---|---|---|---|---|
1 | FairMOT | 67.5 | A Simple Baseline for Multi-Object Tracking | 2020 |
2 | Lif_T | 60.5 | Lifted Disjoint Paths with Application in Multiple Object Tracking | 2020 |
3 | MPNTrack | 58.8 | Learning a Neural Solver for Multiple Object Tracking | 2020 |
4 | DeepMOT | 53.7 | How To Train Your Deep Multi-Object Tracker | 2019 |
5 | JBNOT | 52.6 | Multiple People Tracking using Body and Joint Detections | 2019 |
6 | TNT | 51.9 | Exploit the Connectivity: Multi-Object Tracking with TrackletNet | 2018 |
7 | FWT | 51.3 | Fusion of Head and Full-Body Detectors for Multi-Object Tracking | 2017 |
8 | MOTDT17 | 50.9 | Real-time Multiple People Tracking with Deeply Learned Candidate Selection and Person Re-Identification | 2018 |
<a id="markdown-mot20" name="mot20"></a>
MOT20
Rank | Model | MOTA | Paper | Year |
---|---|---|---|---|
1 | FairMOT | 61.8 | A Simple Baseline for Multi-Object Tracking | 2020 |
2 | UnsupTrack | 53.6 | Simple Unsupervised Multi-Object Tracking | 2020 |
<a id="markdown-toolbox" name="toolbox"></a>
Toolbox
mmtracking: OpenMMLab Video Perception Toolbox. It supports Single Object Tracking (SOT), Multiple Object Tracking (MOT), Video Object Detection (VID) with a unified framework.
<a id="markdown-course" name="course"></a>
Course
link is a good course about multiple object tracking. The course is offered as a Massive Open Online Course (MOOC) on edX.