Memory Enhanced Global-Local Aggregation for Video Object Detection

Chen, Yihong; Cao, Yue; Hu, Han; Wang, Liwei

doi:10.1109/cvpr42600.2020.01035

Cited by 266 publications

(220 citation statements)

References 25 publications

Supporting

Mentioning

217

Contrasting

Order By: Relevance

“…Figure 3 illustrates the framework of our proposed aero engine spark detection model. The model does not follow the pattern of popular video object detectors, which generally make use of candidate boxes from adjacent frames to speed up the current frame detection, or to make up for the missing features of the object in the current frame due to motion blur, vibration of cameras, and so on in order to reduce missed detections [ 32 ]. Differing from such video object detection methods, each result of our model was based on the spatio-temporal features extracted from the spatio-temporal context sequences, rather than candidate boxes of adjacent frames.…”

Section: Methodsmentioning

confidence: 99%

“…We conducted a comparison of three advanced image object detectors, two advanced video object detectors, and SAVADN. Table 2 and Table 3 show the results of the comparison, using three state-of-the-art image object detectors (RetinaNet [ 43 ], YOLOv4 [ 44 ], and YOLOv5) and two state-of-the-art video object detectors (LSTS [ 34 ] and MEGA [ 32 ]). It can be seen that our algorithm significantly outperformed the other algorithms.…”

Section: Methodsmentioning

confidence: 99%

“…Therefore, video object detection has enjoyed great popularity in the field of object detection [ 27 , 28 , 29 ]. Video object detection can be mainly categorized into high-precision approaches [ 30 , 31 , 32 , 33 ] and high-speed approaches [ 34 , 35 , 36 , 37 ]. A high-speed approach makes full use of the temporal context to speed up the process of detection, while high-precision approaches improve the performance by focusing on the temporal context [ 25 ].…”

Section: Introductionmentioning

confidence: 99%

“…A high-speed approach makes full use of the temporal context to speed up the process of detection, while high-precision approaches improve the performance by focusing on the temporal context [ 25 ]. Related models proposed in recent years include FGFA [ 31 ], D&T [ 33 ], MEGA [ 32 ], LSTS [ 34 ], and SELSA [ 30 ]. We found, through experiments, that the state-of-the-art video object detector, MEGA, is also unable to perfectly detect sparks, possibly because it focuses on using the candidate boxes of adjacent frames to guide the current detection.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

SAVSDN: A Scene-Aware Video Spark Detection Network for Aero Engine Intelligent Test

Kou

Zhang

Huang

et al. 2021

Sensors

View full text Add to dashboard Cite

Due to carbon deposits, lean flames, or damaged metal parts, sparks can occur in aero engine chambers. At present, the detection of such sparks deeply depends on laborious manual work. Considering that interference has the same features as sparks, almost all existing object detectors cannot replace humans in carrying out high-precision spark detection. In this paper, we propose a scene-aware spark detection network, consisting of an information fusion-based cascading video codec-image object detector structure, which we name SAVSDN. Unlike video object detectors utilizing candidate boxes from adjacent frames to assist in the current prediction, we find that efforts should be made to extract the spatio-temporal features of adjacent frames to reduce over-detection. Visualization experiments show that SAVSDN can learn the difference in spatio-temporal features between sparks and interference. To solve the problem of a lack of aero engine anomalous spark data, we introduce a method to generate simulated spark images based on the Gaussian function. In addition, we publish the first simulated aero engine spark data set, which we name SAES. In our experiments, SAVSDN far outperformed state-of-the-art detection models for spark detection in terms of five metrics.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Methodsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

SAVSDN: A Scene-Aware Video Spark Detection Network for Aero Engine Intelligent Test

Kou

Zhang

Huang

et al. 2021

Sensors

View full text Add to dashboard Cite

show abstract

“…Deep learning has been increasingly utilized in computer vision tasks recently. Especially the emergence of CNN has resulted in a breakthrough in the research on object detection [ 12 , 13 , 14 , 15 ] and other computer vision tasks [ 16 ]. Object detection methods based on deep learning are classified into two categories: one-stage detection and two-stage detection.…”

Section: Related Workmentioning

confidence: 99%

Multi-Scale Feature Pyramid Network: A Heavily Occluded Pedestrian Detection Network Based on ResNet

Shao

Wang

Yang

et al. 2021

Sensors

View full text Add to dashboard Cite

The existing pedestrian detection algorithms cannot effectively extract features of heavily occluded targets which results in lower detection accuracy. To solve the heavy occlusion in crowds, we propose a multi-scale feature pyramid network based on ResNet (MFPN) to enhance the features of occluded targets and improve the detection accuracy. MFPN includes two modules, namely double feature pyramid network (FPN) integrated with ResNet (DFR) and repulsion loss of minimum (RLM). We propose the double FPN which improves the architecture to further enhance the semantic information and contours of occluded pedestrians, and provide a new way for feature extraction of occluded targets. The features extracted by our network can be more separated and clearer, especially those heavily occluded pedestrians. Repulsion loss is introduced to improve the loss function which can keep predicted boxes away from the ground truths of the unrelated targets. Experiments carried out on the public CrowdHuman dataset, we obtain 90.96% AP which yields the best performance, 5.16% AP gains compared to the FPN-ResNet50 baseline. Compared with the state-of-the-art works, the performance of the pedestrian detection system has been boosted with our method.

show abstract

GPNet: Key Point Generation Auxiliary Network for Object Detection

Shao

Sun

Liu

et al. 2023

Advcd Theory and Sims

View full text Add to dashboard Cite

For existing object detectors, anchor‐based detectors lack global information, while anchor‐free detectors based on key points lack prior position information. The above issues may lead to the imbalance of detection accuracy and shape robustness. In order to alleviate the above contradictions, a new auxiliary network key point generation network (GPNet) is proposed to improve the performance of existing object detectors. Specifically, a series of key points are generated through ground truth (GT) supervision to obtain more global information. These key points generate pseudo boxes (Pbox) with learnable parameters. Pbox has more specific prior information than the manually designed candidate boxes. The scale information of the Pbox is embed into the classification branch to obtain a more appropriate receptive field. In addition, a novel improved strategy for label assignment by combining Pbox and GT to enhance the ability to classify positive and negative samples is proposed. Extensive experiments on multiple dense prediction methods validate the effectiveness of GPNet, with a performance improvement of 1.5 AP over baseline. In particular, with ResNext‐101‐64× 4d‐DCN as the backbone, this method achieves 49.5 AP with single‐scale testing.

show abstract

Memory Enhanced Global-Local Aggregation for Video Object Detection

Cited by 266 publications

References 25 publications

SAVSDN: A Scene-Aware Video Spark Detection Network for Aero Engine Intelligent Test

SAVSDN: A Scene-Aware Video Spark Detection Network for Aero Engine Intelligent Test

Multi-Scale Feature Pyramid Network: A Heavily Occluded Pedestrian Detection Network Based on ResNet

GPNet: Key Point Generation Auxiliary Network for Object Detection

Contact Info

Product

Resources

About