Uncertainty-based Object Detector for Autonomous Driving Embedded Platforms

Choi, Jung-Oh; Chun, Dayoung; Lee, Hyuk-Jae; Kim, Hyun

doi:10.1109/aicas48895.2020.9073907

Cited by 42 publications

(29 citation statements)

References 6 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Compared to YOLOv2 network [34] and its quantized model [40], the proposed network achieves 6.42 higher mAP even though the weight size of the proposed network is approximately onequarter and one-half smaller, respectively. Compared to YOLOv3-tiny [10], Gaussian YOLOv3-tiny [13], and Nayak et al [41], the weight size of the proposed network is approximately one-third larger, but the mAP of the proposed network is much higher than that of YOLOv3-tiny [10], Gaussian YOLOv3-tiny [13], and Nayak et al [41] by 21.42, 15.22, and 26.43, respectively. In particular, through the comparison result with Gaussian YOLOv3-tiny [13], which enhances the accuracy of a small-scale network, it can be seen that effectively reducing the size of a large-scale network is more efficient than improving the performance of a small-scale network.…”

Section: Performance Evaluation On Object Detectionmentioning

confidence: 83%

“…Compared to YOLOv3-tiny [10], Gaussian YOLOv3-tiny [13], and Nayak et al [41], the weight size of the proposed network is approximately one-third larger, but the mAP of the proposed network is much higher than that of YOLOv3-tiny [10], Gaussian YOLOv3-tiny [13], and Nayak et al [41] by 21.42, 15.22, and 26.43, respectively. In particular, through the comparison result with Gaussian YOLOv3-tiny [13], which enhances the accuracy of a small-scale network, it can be seen that effectively reducing the size of a large-scale network is more efficient than improving the performance of a small-scale network. Consequently, the proposed network is superior in terms of the trade-off between accuracy and network size.…”

Section: Performance Evaluation On Object Detectionmentioning

confidence: 83%

“…Weight size (MB) Original YOLOv3 [10] 54.64 32 237 YOLOv2 [34] 48.1 32 194 Zhang et al [40] 48.1 16 97 YOLOv3-tiny [10] 33.1 32 33.79 Gaussian YOLOv3-tiny [13] 39.…”

Section: Methods Map (%) W Bitsmentioning

confidence: 99%

“…Given these advantages, YOLOv3 [10] and YOLOv4 [11] are widely used in domains where both accuracy and speed are important. In particular, it is widely used in embedded platforms, where many recent studies have been conducted [13]. However, even this one-stage object detector is constructed by stacking many layers with high complexity, so the network size is still large, and significant computation is required.…”

Section: Yolo Networkmentioning

confidence: 99%

See 3 more Smart Citations

Zero-Centered Fixed-Point Quantization With Iterative Retraining for Deep Convolutional Neural Network-Based Object Detectors

Kim

2021

IEEE Access

Self Cite

View full text Add to dashboard Cite

In the field of object detection, deep learning has greatly improved accuracy compared to previous algorithms and has been used widely in recent years. However, object detection using deep learning requires many hardware (HW) resources due to the huge computations for high performance, making it very difficult to run real-time on embedded platforms. Therefore, various compression methods have been studied to solve this problem. In particular, quantization methods greatly reduce the computational burden of deep learning by reducing the number of bits used for weights and activation functions in deep learning. However, most of these existing studies targeted only object classification and cannot be applied to object detection. Furthermore, most of the existing quantization studies are based on floating-point operations, which requires additional effort when implementing HW accelerators. This paper proposes an HW-friendly fixed-point-based quantization method that can also be applied to object detection. In the proposed method, the center of the weight distribution is adjusted to zero by subtracting the mean of weight parameters before quantization, and the retraining process is iteratively applied to minimize the accuracy drop caused by quantization. Furthermore, while applying the proposed method to object detection, performance degradation is minimized by considering the minimum and maximum values of weight parameters of deep learning networks. When applying the proposed quantization method to representative one-stage object detectors, You Only Look Once v3 and v4 (YOLOv3 and YOLOv4), detection accuracy similar to the original networks (i.e., YOLOv3 and YOLOv4) with a single-precision floating-point format (32-bit) is maintained despite expressing weights with only about 20% of the bits compared to a singleprecision floating-point format in COCO dataset. INDEX TERMS Convolutional neural network, deep neural network, fixed-point quantization, network compression, object detector, YOLOv3, YOLOv4.

show abstract

Section: Performance Evaluation On Object Detectionmentioning

confidence: 83%

Section: Performance Evaluation On Object Detectionmentioning

confidence: 83%

“…Weight size (MB) Original YOLOv3 [10] 54.64 32 237 YOLOv2 [34] 48.1 32 194 Zhang et al [40] 48.1 16 97 YOLOv3-tiny [10] 33.1 32 33.79 Gaussian YOLOv3-tiny [13] 39.…”

Section: Methods Map (%) W Bitsmentioning

confidence: 99%

Section: Yolo Networkmentioning

confidence: 99%

See 2 more Smart Citations

Zero-Centered Fixed-Point Quantization With Iterative Retraining for Deep Convolutional Neural Network-Based Object Detectors

Kim

2021

IEEE Access

Self Cite

View full text Add to dashboard Cite

show abstract

“…Zhang et al [55] proposed a new platform called Caffeine for FPGA hardware, and the execution speed was improved 9 29 in performance. [25] and [6] focus their research on vehicle detection which has a key impact in autonomous driving field. Both, based their development in YOLOv3-based algorithm achieving improvements in the accuracy of at least 7% more.…”

Section: Convolutional Neural Network Improvementsmentioning

confidence: 99%

A dynamic discarding technique to increase speed and preserve accuracy for YOLOv3

Martinez‐Alpiste

Golkarnarenji

Wang

et al. 2021

Neural Comput & Applic

View full text Add to dashboard Cite

This paper proposes an acceleration technique to minimise the unnecessary operations on a state-of-the-art machine learning model and thus to improve the processing speed while maintaining the accuracy. After the study of the main bottlenecks that negatively affect the performance of convolutional neural networks, this paper designs and implements a discarding technique for YOLOv3-based algorithms to increase the speed and maintain accuracy. After applying the discarding technique, YOLOv3 can achieve a 22% of improvement in terms of speed. Moreover, the results of this new discarding technique were tested on Tiny-YOLOv3 with three output layers on an autonomous vehicle for pedestrian detection and it achieved an improvement of 48.7% in speed. The dynamic discarding technique just needs one training process to create the model and thus execute the approach, which preserves accuracy. The improved detector based on the discarding technique is able to readily alert the operator of the autonomous vehicle to take the emergency brake of the vehicle in order to avoid collision and consequently save lives.

show abstract

Uncertainty Quantification for Object Detection: Output- and Gradient-Based Approaches

Riedlinger

Schubert

Kahl

et al. 2022

Deep Neural Networks and Data for Automated Driving

View full text Add to dashboard Cite

Safety-critical applications of deep neural networks require reliable confidence estimation methods with high predictive power. However, evaluating and comparing different methods for uncertainty quantification is oftentimes highly context-dependent. In this chapter, we introduce flexible evaluation protocols which are applicable to a wide range of tasks with an emphasis on object detection. In this light, we investigate uncertainty metrics based on the network output, as well as metrics based on a learning gradient, both of which significantly outperform the confidence score of the network. While output-based uncertainty is produced by post-processing steps and is computationally efficient, gradient-based uncertainty, in principle, allows for localization of uncertainty within the network architecture. We show that both sources of uncertainty are mutually non-redundant and can be combined beneficially. Furthermore, we show direct applications of uncertainty quantification by improving detection accuracy.

show abstract

Uncertainty-based Object Detector for Autonomous Driving Embedded Platforms

Cited by 42 publications

References 6 publications

Zero-Centered Fixed-Point Quantization With Iterative Retraining for Deep Convolutional Neural Network-Based Object Detectors

Zero-Centered Fixed-Point Quantization With Iterative Retraining for Deep Convolutional Neural Network-Based Object Detectors

A dynamic discarding technique to increase speed and preserve accuracy for YOLOv3

Uncertainty Quantification for Object Detection: Output- and Gradient-Based Approaches

Contact Info

Product

Resources

About