Memristive DeepLab: A hardware friendly deep CNN for semantic segmentation

Zhang, Lin; Hu, Xiaofang; Zhou, Yue; Zhou, Guangdong; Duan, Shukai

doi:10.1016/j.neucom.2021.04.061

Cited by 20 publications

(6 citation statements)

References 32 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In principle, pixels with similar distances will be assigned the same semantic labels as much as possible, and pixels with obvious differences need to be assigned different labels. The evaluation index of “distance” defined by color difference and spatial relative distance ensures that the image can be accurately cut at the edge with a large gradient to a certain extent [ 25 , 26 ]. Different from the ordinary conditional random field, the bivariate term in the fully connected conditional random field expresses the correlation between each pixel and all other pixels in the image.…”

Section: Image Semantic Segmentation Based On Deep Fusion Network Com...mentioning

confidence: 99%

Image Semantic Segmentation Method Based on Deep Fusion Network and Conditional Random Field

Wang

Yang

2022

Computational Intelligence and Neuroscience

View full text Add to dashboard Cite

Aiming at the problems of missing points and wrong points in image semantic segmentation under complex background and small target, an image semantic segmentation method based on the fully convolution neural network and conditional random field is proposed. First, the deconvolution fusion structure is added to the fully convolution neural network to build a deep fusion network. The multiscale features are automatically obtained through the deep fusion network, and the shallow detail information and deep semantic information are fused to improve the processing accuracy of image rough segmentation. Then, the bivariate potential function of the conditional random field is optimized based on the convolution neural network, and it is used for image fine segmentation to obtain the final image segmentation result. Finally, the proposed method is experimentally analyzed based on the Cityscapes dataset. The results show that the proposed method can achieve accurate image segmentation, and the area under the segmentation curve of the overall size target is 93.6%, which is better than other methods.

show abstract

Section: Image Semantic Segmentation Based On Deep Fusion Network Com...mentioning

confidence: 99%

Image Semantic Segmentation Method Based on Deep Fusion Network and Conditional Random Field

Wang

Yang

2022

Computational Intelligence and Neuroscience

View full text Add to dashboard Cite

show abstract

“…The limited storage and computing resources of FPGA and the innovation of network models have challenged the generality and energy efficiency of neural networks. Many current neural network accelerators [ 9 , 10 ] improve speed and performance by incorporating lightweight network models and new hardware architectures. The Roofline model [ 11 ] can improve the overall system throughput by increasing the data transmission bandwidth and number of unit memory access calculations under different constraints.…”

Section: Introductionmentioning

confidence: 99%

Efficient Binary Weight Convolutional Network Accelerator for Speech Recognition

Guo

Deng

et al. 2023

Sensors

View full text Add to dashboard Cite

Speech recognition has progressed tremendously in the area of artificial intelligence (AI). However, the performance of the real-time offline Chinese speech recognition neural network accelerator for edge AI needs to be improved. This paper proposes a configurable convolutional neural network accelerator based on a lightweight speech recognition model, which can dramatically reduce hardware resource consumption while guaranteeing an acceptable error rate. For convolutional layers, the weights are binarized to reduce the number of model parameters and improve computational and storage efficiency. A multichannel shared computation (MCSC) architecture is proposed to maximize the reuse of weight and feature map data. The binary weight-sharing processing engine (PE) is designed to avoid limiting the number of multipliers. A custom instruction set is established according to the variable length of voice input to configure parameters for adapting to different network structures. Finally, the ping-pong storage method is used when the feature map is an input. We implemented this accelerator on Xilinx ZYNQ XC7Z035 under the working frequency of 150 MHz. The processing time for 2.24 s and 8 s of speech was 69.8 ms and 189.51 ms, respectively, and the convolution performance reached 35.66 GOPS/W. Compared with other computing platforms, accelerators perform better in terms of energy efficiency, power consumption and hardware resource consumption.

show abstract

“…For traditional semantic segmentation such as gray segmentation and conditional random fields, the underlying features of the image are usually used to divide the region of the image, and its segmentation accuracy needs to be further improved. At present, with the development of convolutional neural network algorithm (CNN) and its application in semantic segmentation, a large number of semantic segmentation models based on deep learning have been proposed, which can solve the problem of difficult feature selection in traditional semantic segmentation [7].…”

Section: Introductionmentioning

confidence: 99%

BiSeNet-oriented context attention model for image semantic segmentation

Teng

Qiao

2022

COMSIS J

View full text Add to dashboard Cite

When the traditional semantic segmentation model is adopted, the different feature importance of feature maps is ignored in the feature extraction stage, which results in the detail loss, and affects the segmentation effect. In this paper, we propose a BiSeNet-oriented context attention model for image semantic segmentation. In the BiSeNet, the spatial path is utilized to extract more low-level features to solve the problem of information loss in deep network layers. Context attention mechanism is used to mine high-level implied semantic features of images. Mean while, the focus loss is used as the loss function to improve the final segmentation effect by reducing the internal weighting. Finally, we conduct experiments on open data sets, and the results show that pixel accuracy, average pixel accuracy, and aver age Intersection-over-Union are greatly improved compared with other state-of-the art semantic segmentation models. It effectively improves the accuracy of feature extraction, reduces the loss of feature details, and improves the final segmentation effect.

show abstract

Memristive DeepLab: A hardware friendly deep CNN for semantic segmentation

Cited by 20 publications

References 32 publications

Image Semantic Segmentation Method Based on Deep Fusion Network and Conditional Random Field

Image Semantic Segmentation Method Based on Deep Fusion Network and Conditional Random Field

Efficient Binary Weight Convolutional Network Accelerator for Speech Recognition

BiSeNet-oriented context attention model for image semantic segmentation

Contact Info

Product

Resources

About