Overcoming Classifier Imbalance for Long-Tail Object Detection With Balanced Group Softmax

Li, Yu; Wang, Tao; Kang, Bingyi; Tang, Sheng; Wang, Chunfeng; Li, Jintao; Feng, Jiashi

doi:10.1109/cvpr42600.2020.01100

Cited by 217 publications

(199 citation statements)

References 21 publications

Supporting

Mentioning

197

Contrasting

Order By: Relevance

“…The Softmax is also known as multinomial regression and can be used for mutually exclusive multiclass classification. It has been widely used in deep learning, yielding excellent performance [56,57]. The mathematical definition of Softmax is given as…”

Section: Deepmentioning

confidence: 99%

Enhanced Intelligent Smart Home Control and Security System Based on Deep Learning Model

Taiwo

Ezugwu

Oyelade

et al. 2022

Wireless Communications and Mobile Computing

View full text Add to dashboard Cite

Security of lives and properties is highly important for enhanced quality living. Smart home automation and its application have received much progress towards convenience, comfort, safety, and home security. With the advances in technology and the Internet of Things (IoT), the home environment has witnessed an improved remote control of appliances, monitoring, and home security over the internet. Several home automation systems have been developed to monitor movements in the home and report to the user. Existing home automation systems detect motion and have surveillance for home security. However, the logical aspect of averting unnecessary or fake notifications is still a major area of challenge. Intelligent response and monitoring make smart home automation efficient. This work presents an intelligent home automation system for controlling home appliances, monitoring environmental factors, and detecting movement in the home and its surroundings. A deep learning model is proposed for motion recognition and classification based on the detected movement patterns. Using a deep learning model, an algorithm is developed to enhance the smart home automation system for intruder detection and forestall the occurrence of false alarms. A human detected by the surveillance camera is classified as an intruder or home occupant based on his walking pattern. The proposed method’s prototype was implemented using an ESP32 camera for surveillance, a PIR motion sensor, an ESP8266 development board, a 5 V four-channel relay module, and a DHT11 temperature and humidity sensor. The environmental conditions measured were evaluated using a mathematical model for the response time to effectively show the accuracy of the DHT sensor for weather monitoring and future prediction. An experimental analysis of human motion patterns was performed using the CNN model to evaluate the classification for the detection of humans. The CNN classification model gave an accuracy of 99.8%.

show abstract

Section: Deepmentioning

confidence: 99%

Enhanced Intelligent Smart Home Control and Security System Based on Deep Learning Model

Taiwo

Ezugwu

Oyelade

et al. 2022

Wireless Communications and Mobile Computing

View full text Add to dashboard Cite

show abstract

“…After the release of LVIS dataset, To further tackle the class imbalance issue, Tan et al propose EQL [53] that ignores discouraging gradients for negative samples. Li et al propose balanced group softmax [35] that balances the classifiers through group-wise training. In this paper, we propose bucketed class re-weighting that adaptively partitions the categories into different magnitudes, and use sigmoid function to adjust their weights.…”

Section: Balanced Learning For Instance Recognitionmentioning

confidence: 99%

Towards Balanced Learning for Instance Recognition

Pang

Chen²,

Li³

et al. 2021

Int J Comput Vis

View full text Add to dashboard Cite

Instance recognition is rapidly advanced along with the developments of various deep convolutional neural networks. Compared to the architectures of networks, the training process, which is also crucial to the success of detectors, has received relatively less attention. In this work, we carefully revisit the standard training practice of detectors, and find that the detection performance is often limited by the imbalance during the training process, which generally consists in three levels -sample level, feature level, and objective level. To mitigate the adverse effects caused thereby, we propose Libra R-CNN, a simple yet effective framework towards balanced learning for instance recognition. It integrates IoU-balanced sampling, balanced feature pyramid, and objective re-weighting, respectively for reducing the imbalance at sample, feature, and objective level. Extensive experiments conducted on MS COCO, LVIS and Pascal VOC datasets prove the effectiveness of the overall balanced design.

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

Overcoming Classifier Imbalance for Long-Tail Object Detection With Balanced Group Softmax

Cited by 217 publications

References 21 publications

Enhanced Intelligent Smart Home Control and Security System Based on Deep Learning Model

Enhanced Intelligent Smart Home Control and Security System Based on Deep Learning Model

Towards Balanced Learning for Instance Recognition

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

Contact Info

Product

Resources

About