• Laser & Optoelectronics Progress
  • Vol. 57, Issue 4, 041011 (2020)
Shiqing Huang1, Ruilin Bai1、*, and Gaoe Qin2
Author Affiliations
  • 1Key Laboratory of Advanced Process Control for Light Industry (Ministry of Education), Jiangnan University, Wuxi, Jiangsu 214122, China
  • 2Xinje Electronic Co., Ltd, Wuxi, Jiangsu 214122, China
  • show less
    DOI: 10.3788/LOP57.041011 Cite this Article Set citation alerts
    Shiqing Huang, Ruilin Bai, Gaoe Qin. Method of Convolutional Neural Network Model Pruning Based on Gray Correlation Analysis[J]. Laser & Optoelectronics Progress, 2020, 57(4): 041011 Copy Citation Text show less
    Framework of model pruning
    Fig. 1. Framework of model pruning
    Schematic of convolutional neutral network
    Fig. 2. Schematic of convolutional neutral network
    Schematic of pruning of residual network
    Fig. 3. Schematic of pruning of residual network
    Schematic of pruning quantity of each layer
    Fig. 4. Schematic of pruning quantity of each layer
    Precision change in pruning process
    Fig. 5. Precision change in pruning process
    Comparison of pruning accuracy changes of various methods. (a) Comparison of various methods on VGG-16; (b) comparison of various methods on ResNet-50; (c) comparison of various methods on AlexNet
    Fig. 6. Comparison of pruning accuracy changes of various methods. (a) Comparison of various methods on VGG-16; (b) comparison of various methods on ResNet-50; (c) comparison of various methods on AlexNet
    ModelPruning ratio /%Accuracy /%Acceleration factoron 1080TiAcceleration factoron TX2Size /MB
    VGG-16093.143.5
    VGG-16-pruned-A4092.01.36×1.83 ×17.6
    VGG-16-pruned-B8091.82.7×2.8×3.2
    ResNet-50088.474.3
    ResNet-50-pruned-A4085.41.46×1.6×43.8
    ResNet-50-pruned-B8079.33.4×1.9×8.9
    AlexNet078.214.2
    AlexNet-pruned-A4071.51.6×1.9×6.4
    AlexNet-pruned-B8054.23.1×3.4 ×1.8
    Table 1. Overall experimental results
    Pruningratio /%Accuracy(Top-5) /%Variation ofaccuracy /%Accelerationfactor on 1080TiAccelerationfactor on TX2Time on 1080Tiand TX2 /msSize /MB
    078.201.001.0032,7414.2
    2075.2-3.01.281.4225,529.6
    4071.5-6.71.601.9020,396.4
    6068.7-9.52.102.3015,322.9
    8054.2-24.03.103.4010,211.8
    Table 2. Accuracy and acceleration effect of AlexNet
    Pruningratio /%Accuracy(Top-5) /%Variation ofaccuracy /%Accelerationfactor on 1080TiAccelerationfactor on TX2Time on 1080Tiand TX2 /msSize /MB
    088.401.001.079,24874.3
    2086.5-1.91.171.467,17867.6
    4085.4-3.01.461.654,15443.8
    6083.1-5.31.971.840,13625.3
    8079.3-9.13.401.923,1288.9
    Table 3. Accuracy and acceleration effect of ResNet-50
    Shiqing Huang, Ruilin Bai, Gaoe Qin. Method of Convolutional Neural Network Model Pruning Based on Gray Correlation Analysis[J]. Laser & Optoelectronics Progress, 2020, 57(4): 041011
    Download Citation