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

Abstract

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.

Exploration and Strategy Analysis of Mental Health Education for Students in Sports Majors in the Era of Artificial Intelligence

“…The number of iterations of the CNN model obtained here is basically consistent with the number of iterations obtained by Kim and Kim (2021) in CNN research ( Kim and Kim, 2021 ). It shows that applying the CNN model to mental state early warning is feasible.…”

Research on Navigation and Positioning of Electric Inspection Robot Based on Improved CNN Algorithm

“…At the same time, the video will be transmitted to the Alibaba cloud platform for backup, which can be retrieved by the inspectors at any time. It can also directly follow the real-time video and conduct video remote detection synchronously with the machine [11,12]. Compared with the early inspection robot, the advantage of this power inspection robot is to increase the image detection function of convolutional neural network.…”

Quantization of Weights of Neural Networks with Negligible Decreasing of Prediction Accuracy

“…In order to compare performance of the designed USQ, we employ the quantizer (the uniform one) used in fixed-point format representations [6,14], conducting the analysis for R = 9 bps. In particular, the generated codeword of baseline quantizer consists of one bit reserved for sign (s = 1), n bits reserved for integer part and m bits reserved for fractional We can see in Figure 3 that, as expected, higher SQNR values is obtained as the bit rate R increases.…”

“…In order to compare performance of the designed USQ, we employ the quantizer (the uniform one) used in fixed-point format representations [6,14], quantization noise ratio (SQNR, g SQNR and ov SQNR ) versus ρ for the proposed USQ (for R = 9 bps).…”

“…Finally, we provide the results in case of the baseline quantizer approach discussed in [6,14], taking into account R = 9 bps. SQNR versus ρ, obtained from real data (weights), can be found in Figure 9, where good agreement with theoretical results in Figure 5 is observed.…”