Chained-Tracker: Chaining Paired Attentive Regression Results for End-to-End Joint Multiple-Object Detection and Tracking

Peng, Jinlong; Wang, Chang‐An; Wan, Fangbin; Wu, Yang; Wang, Yabiao; Tai, Ying; Wang, Chengjie; Li, Jilin; Huang, Feiyue; Fu, Yanwei

doi:10.1007/978-3-030-58548-8_9

Cited by 289 publications

(143 citation statements)

References 36 publications

Supporting

Mentioning

143

Contrasting

Order By: Relevance

“…For a fair comparison, we only compare methods that use a private detector. As shown in Table 1, our weakly-supervised MOT model shows significant results compared to those of the state-of-the-art fully supervised object trackers [5,6,7,8]. Furthermore, our weakly-supervised model even outperforms CenterTrack by 0.6% in terms of MOTA, which is significant.…”

Section: Results On the Mot Benchmarkmentioning

confidence: 81%

“…We used the DLA-34 network [24] as a backbone architecture, and optimized the whole framework with Adam optimizer [25]. We set the learning rate to 1.56×10 −5 (decreased [20], IDF1 [21] and ID switches (IDs) [22], of TubeTK [5], GSDT [6], CTracker [7], and CenterTrack [8] are from the leaderboard of MOT17 Challenge [23]. We have officially released our results on the MOT challenge website as MW MOT.…”

Section: Methodsmentioning

confidence: 99%

“…With the advancement of convolutional neural networks (CNNs) and their feature extraction ability, the performance of MOT methods has greatly improved. However, most MOT methods [4,5,6,7,8] adopt supervised learning schemes, where the bounding box, as well as the ID of each object for each frame, must be provided as the ground truth during training. It especially requires significant time and effort to correctly label the training data since the object ID of a few tens to thousands of detected objects must be labeled for each frame of lengthy video sequences (e.g.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Weakly-Supervised Multiple Object Tracking Via A Masked Center Point Warping Loss

Soon-Ock

Shim

Park

et al. 2021

2021 IEEE International Conference on Image Processing (ICIP)

View full text Add to dashboard Cite

Multiple object tracking (MOT), a popular subject in computer vision with broad application areas, aims to detect and track multiple objects across an input video. However, recent learning-based MOT methods require strong supervision on both the bounding box and the ID of each object for every frame used during training, which induces a heightened cost for obtaining labeled data. In this paper, we propose a weakly-supervised MOT framework that enables the accurate tracking of multiple objects while being trained without object ID ground truth labels. Our model is trained only with the bounding box information with a novel masked warping loss that drives the network to indirectly learn how to track objects through a video. Specifically, valid object center points in the current frame are warped with the predicted offset vector and enforced to be equal to the valid object center points in the previous frame. With this approach, we obtain an MOT accuracy on par with those of the state-of-the-art fully supervised MOT models, which use both the bounding boxes and object ID as ground truth labels, on the MOT17 dataset.

show abstract

Section: Results On the Mot Benchmarkmentioning

confidence: 81%

Section: Methodsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Weakly-Supervised Multiple Object Tracking Via A Masked Center Point Warping Loss

Soon-Ock

Shim

Park

et al. 2021

2021 IEEE International Conference on Image Processing (ICIP)

View full text Add to dashboard Cite

show abstract

“…The model uses the principle of minimizing heat maps offset for identity matching. Chained-Tracker [13] uses a chained tracking model. It utilizes the Siamese network to detect the two frames object and only employs the improved IoU matching for data association.…”

Section: Related Workmentioning

confidence: 99%

“…Therefore, such model directly outputs the tracking result without additional data association or tracking processing on the network information. End-to-end model typically employs additional neural networks [15], single object tracking (SOT) networks [25] or improved intersection over union (IoU) matching [13] [14], heatmap offsets matching [12] and so on for data association. Although the model has a concise tracking paradigm, they do not have advantage in the expansion of network structure and the balance in accuracy and speed.…”

Section: Introductionmentioning

confidence: 99%

Appearance Guidance Attention for Multi-Object Tracking

et al. 2021

View full text Add to dashboard Cite

Appearance information is one of the most important matching indicators for multi-object data association. In tracking by detection model, appearance information and detection information are usually integrated in the same sub-network for learning and output. This phenomenon will result in the appearance embedding vectors to be coupled to the network inference method during learning stage. As a result, the appearance embedding vectors contains too much background information and affects the accuracy of data association. Based on the non-end-to-end tracking model, we design an appearance guidance attention module for the appearance extraction branch. This module can effectively strengthen the network's learning of the object visual appearance features and reduce the attention in the learning of background features. Finally, we utilize the appearance embedding vectors that decoupled from the inference method as the input of the back-end tracker and perform data association. The proposed method is tested on the MOT16 and MOT17 datasets. Experiments show that the proposed method provides more high-quality appearance representation information for the back-end tracker and the tracking performance on two datasets is better than other comparison models. At the same time, our model can reach 24.9FPS on a single 1080Ti GPU.

show abstract

Attention-Enabled Object Detection to Improve One-Stage Tracker

Madan

Nasrollahi

Moeslund

2021

Lecture Notes in Networks and Systems

View full text Add to dashboard Cite

State-of-the-art (SoTA) detection-based tracking methods mostly accomplish the detection and the identification feature learning tasks separately. Only a few efforts include the joint learning of detection and identification features. This work proposes two novel one-stage trackers by introducing implicit and explicit attention to the tracking research topic. For our tracking system based on implicit attention, we further introduce a novel fusion of feature maps combining information from different abstraction levels. For our tracking system based on explicit attention, we introduce utilization of an additional auxiliary function. These systems outperform the SoTA tracking systems in terms of MOTP (Multi-Object Tracking Precision) and IDF1 score when evaluated on public benchmark datasets including MOT15, MOT16, and MOT17. High MOTP score indicates precise detection of bounding boxes of objects, while high IDF1 score indicates accurate ID detections, which is very crucial for surveillance and security systems. Therefore, proposed systems are good choice for event-detections in surveillance feeds as we are capable of detecting correct ID and precise location.

show abstract

Chained-Tracker: Chaining Paired Attentive Regression Results for End-to-End Joint Multiple-Object Detection and Tracking

Cited by 289 publications

References 36 publications

Weakly-Supervised Multiple Object Tracking Via A Masked Center Point Warping Loss

Weakly-Supervised Multiple Object Tracking Via A Masked Center Point Warping Loss

Appearance Guidance Attention for Multi-Object Tracking

Attention-Enabled Object Detection to Improve One-Stage Tracker

Contact Info

Product

Resources

About