Jian Xu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Jian Xu has a total of 7 co-authors. Here are the top co-authors by number of shared publications:
- Weijun Li
Chinese Academy of Sciences
3 shared publications - Lina Yu
AnnLab, Institute of Semiconductors, Chinese Academy of Sciences, Beijing 100083, China
2 shared publications - Ming Zhang
AnnLab, Institute of Semiconductors, Chinese Academy of Sciences, Beijing 100083, China
2 shared publications - Xin Ning
Institute of Semiconductors, Chinese Academy of Sciences, Beijing, China
2 shared publications - Jinzhong He
AnnLab, Institute of Semiconductors, Chinese Academy of Sciences, Beijing 100083, China
1 shared publication - Liping Zhang
Institute of Semiconductors, Chinese Academy of Sciences, Beijing, China
1 shared publication
Latest Publications
- Open Access
- 12 March 2024
- Published by MDPI AG in Electronics
- Vol. 13 (6) , 1050
Optimizing CNN Hardware Acceleration with Configurable Vector Units and Feature Layout Strategies
Convolutional neural network (CNN) hardware acceleration is critical to improve the performance and facilitate the deployment of CNNs in edge applications. Due to its efficiency and simplicity, channel group parallelism has become a popular method for CNN hardware acceleration. However, when processing data involving small channels, there will be a mismatch between feature data and computing units, resulting in a low utilization of the computing units. When processing the middle layer of the convolutional neural network, the mismatch between the feature-usage order and the feature-loading order leads to a low input feature cache hit rate. To address these challenges, this paper proposes an innovative method inspired by data reordering technology, aiming to achieve CNN hardware acceleration that reuses the same multiplier resources. This method focuses on transforming the hardware acceleration process into feature organization, feature block scheduling and allocation, and feature calculation subtasks to ensure the efficient mapping of continuous loading and the calculation of feature data. Specifically, this paper introduces a convolutional algorithm mapping strategy and a configurable vector operation unit to enhance multiplier utilization for different feature map sizes and channel numbers. In addition, an off-chip address mapping and on-chip cache management mechanism is proposed to effectively improve the feature access efficiency and on-chip feature cache hit rate. Furthermore, a configurable feature block scheduling policy is proposed to strike a balance between weight reuse and feature writeback pressure. Experimental results demonstrate the effectiveness of this method. When using 512 multipliers and accelerating VGG16 at 100 MHz, the actual computing performance reaches 102.3 giga operations per second (GOPS). Compared with other CNN hardware acceleration methods, the average computing array utilization is as high as 99.88% and the computing density is higher.
- 12 March 2024
- Published by MDPI AG in Electronics
- Vol. 13 (6) , 1050
Open AccessOptimizing CNN Hardware Acceleration with Configurable Vector Units and Feature Layout Strategies
Convolutional neural network (CNN) hardware acceleration is critical to improve the performance and facilitate the deployment of CNNs in edge applications. Due to its efficiency and simplicity, channel group parallelism has become a popular method for CNN hardware acceleration. However, when processing data involving small channels, there will be a mismatch between feature data and computing units, resulting in a low utilization of the computing units. When processing the middle layer of the convolutional neural network, the mismatch between the feature-usage order and the feature-loading order leads to a low input feature cache hit rate. To address these challenges, this paper proposes an innovative method inspired by data reordering technology, aiming to achieve CNN hardware acceleration that reuses the same multiplier resources. This method focuses on transforming the hardware acceleration process into feature organization, feature block scheduling and allocation, and feature calculation subtasks to ensure the efficient mapping of continuous loading and the calculation of feature data. Specifically, this paper introduces a convolutional algorithm mapping strategy and a configurable vector operation unit to enhance multiplier utilization for different feature map sizes and channel numbers. In addition, an off-chip address mapping and on-chip cache management mechanism is proposed to effectively improve the feature access efficiency and on-chip feature cache hit rate. Furthermore, a configurable feature block scheduling policy is proposed to strike a balance between weight reuse and feature writeback pressure. Experimental results demonstrate the effectiveness of this method. When using 512 multipliers and accelerating VGG16 at 100 MHz, the actual computing performance reaches 102.3 giga operations per second (GOPS). Compared with other CNN hardware acceleration methods, the average computing array utilization is as high as 99.88% and the computing density is higher.
- Open Access
- 12 December 2023
- Published by MDPI AG in Electronics
- Vol. 12 (24) , 4972
DiffQuant: Reducing Compression Difference for Neural Network Quantization
Deep neural network quantization is a widely used method in the deployment of mobile or edge devices to effectively reduce memory overhead and speed up inference. However, quantization inevitably leads to a reduction in the performance and equivalence of models. Moreover, access to labeled datasets is often denied as they are considered valuable assets for companies or institutes. Consequently, performing quantization training becomes challenging without sufficient labeled datasets. To address these issues, we propose a novel quantization pipeline named DiffQuant, which can perform quantization training using unlabeled datasets. The pipeline includes two cores: the compression difference (CD) and model compression loss (MCL). The CD can measure the degree of equivalence loss between the full-precision and quantized models, and the MCL supports fine-tuning the quantized models using unlabeled data. In addition, we design a quantization training scheme that allows the quantization of both the batch normalization (BN) layer and the bias. Experimental results show that our method outperforms state-of-the-art methods on ResNet18/34/50 networks, maintaining performance with a reduced CD. We achieve Top-1 accuracies of 70.08%, 74.11%, and 76.16% on the ImageNet dataset for the 8-bit quantized ResNet18/34/50 models and reduce the gap to 0.55%, 0.61%, and 0.71% with the full-precision network, respectively. We achieve CD values of only 7.45%, 7.48%, and 8.52%, which allows DiffQuant to further exploit the potential of quantization.
- 12 December 2023
- Published by MDPI AG in Electronics
- Vol. 12 (24) , 4972
Open AccessDiffQuant: Reducing Compression Difference for Neural Network Quantization
Deep neural network quantization is a widely used method in the deployment of mobile or edge devices to effectively reduce memory overhead and speed up inference. However, quantization inevitably leads to a reduction in the performance and equivalence of models. Moreover, access to labeled datasets is often denied as they are considered valuable assets for companies or institutes. Consequently, performing quantization training becomes challenging without sufficient labeled datasets. To address these issues, we propose a novel quantization pipeline named DiffQuant, which can perform quantization training using unlabeled datasets. The pipeline includes two cores: the compression difference (CD) and model compression loss (MCL). The CD can measure the degree of equivalence loss between the full-precision and quantized models, and the MCL supports fine-tuning the quantized models using unlabeled data. In addition, we design a quantization training scheme that allows the quantization of both the batch normalization (BN) layer and the bias. Experimental results show that our method outperforms state-of-the-art methods on ResNet18/34/50 networks, maintaining performance with a reduced CD. We achieve Top-1 accuracies of 70.08%, 74.11%, and 76.16% on the ImageNet dataset for the 8-bit quantized ResNet18/34/50 models and reduce the gap to 0.55%, 0.61%, and 0.71% with the full-precision network, respectively. We achieve CD values of only 7.45%, 7.48%, and 8.52%, which allows DiffQuant to further exploit the potential of quantization.
- 20 February 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computational Social Systems
- Vol. 11 (1) , 1440-1450
A Recognizable Expression Line Portrait Synthesis Method in Portrait Rendering Robot
An artistic line portrait robot can generate, process, and draw line portraits. Compared to real face images, line portraits lose some recognizable information. Maintaining recognizability during the process of expression edition of line portraits is an important challenge for artistic portrait robots. A recognizable expression line portrait synthesis method based on a triangle coordinate system (TCS) is proposed. First, based on public facial expression databases [JAFFE, Oulu CASIA, RaFD, and Cohn-Kanade (CK)], by studying the feature deviations between different expressions of the same person, an expression deformation constraint criterion (EDCC) that is conducive to maintaining recognizable features is proposed. Then, by comparing features between the source line portrait and reference expression portrait, the expression features are calculated. Finally, under the EDCC, based on expression features, a recognizable expression line portrait is generated through image topological deformation based on TCS. In addition, we can synthesize different degrees of expression line portraits. On the public face datasets (FHHQ, CelebA-HQ, and CK), we implemented qualitative and quantitative contrast experiments. Experimental results demonstrate that this method can automatically synthesize an expression line portrait with reference expression, where the expression degree of the reference expression is controllable, and the generated expression portrait still has high recognizability. The expression samples generated by the proposed method are used for face authentication on the CK dataset, and only 0.22% of the samples fail to pass the authentication.
- 20 February 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computational Social Systems
- Vol. 11 (1) , 1440-1450
A Recognizable Expression Line Portrait Synthesis Method in Portrait Rendering Robot
An artistic line portrait robot can generate, process, and draw line portraits. Compared to real face images, line portraits lose some recognizable information. Maintaining recognizability during the process of expression edition of line portraits is an important challenge for artistic portrait robots. A recognizable expression line portrait synthesis method based on a triangle coordinate system (TCS) is proposed. First, based on public facial expression databases [JAFFE, Oulu CASIA, RaFD, and Cohn-Kanade (CK)], by studying the feature deviations between different expressions of the same person, an expression deformation constraint criterion (EDCC) that is conducive to maintaining recognizable features is proposed. Then, by comparing features between the source line portrait and reference expression portrait, the expression features are calculated. Finally, under the EDCC, based on expression features, a recognizable expression line portrait is generated through image topological deformation based on TCS. In addition, we can synthesize different degrees of expression line portraits. On the public face datasets (FHHQ, CelebA-HQ, and CK), we implemented qualitative and quantitative contrast experiments. Experimental results demonstrate that this method can automatically synthesize an expression line portrait with reference expression, where the expression degree of the reference expression is controllable, and the generated expression portrait still has high recognizability. The expression samples generated by the proposed method are used for face authentication on the CK dataset, and only 0.22% of the samples fail to pass the authentication.