项目作者: YeongHyeon

项目描述 :
TensorFlow implementation of "Aggregated Residual Transformations for Deep Neural Networks"
高级语言: Python
项目地址: git://github.com/YeongHyeon/ResNeXt-TF2.git
创建时间: 2020-02-25T07:51:12Z
项目社区:https://github.com/YeongHyeon/ResNeXt-TF2

开源协议:MIT License

下载


[TensorFlow 2] Aggregated Residual Transformations for Deep Neural Networks

TensorFlow implementation of “Aggregated Residual Transformations for Deep Neural Networks”

ResNet-TF2
WideResNet(WRN)-TF2
ResNet-with-LRWarmUp-TF2
ResNet-with-SGDR-TF2

Concept



The three ways for construct ResNeXt block [1].


Performance

Indicator Value
Accuracy 0.99370
Precision 0.99371
Recall 0.99364
F1-Score 0.99367
  1. Confusion Matrix
  2. [[ 977 0 1 0 0 0 0 0 2 0]
  3. [ 0 1129 2 0 0 0 1 2 1 0]
  4. [ 0 1 1026 0 1 0 0 3 1 0]
  5. [ 0 0 2 1007 0 1 0 0 0 0]
  6. [ 0 0 0 0 976 0 1 0 0 5]
  7. [ 1 0 0 4 0 883 2 0 0 2]
  8. [ 1 1 0 0 1 1 953 0 1 0]
  9. [ 0 1 2 0 0 0 0 1024 1 0]
  10. [ 2 0 2 1 0 0 0 2 965 2]
  11. [ 0 1 0 0 4 2 0 3 2 997]]
  12. Class-0 | Precision: 0.99592, Recall: 0.99694, F1-Score: 0.99643
  13. Class-1 | Precision: 0.99647, Recall: 0.99471, F1-Score: 0.99559
  14. Class-2 | Precision: 0.99130, Recall: 0.99419, F1-Score: 0.99274
  15. Class-3 | Precision: 0.99506, Recall: 0.99703, F1-Score: 0.99604
  16. Class-4 | Precision: 0.99389, Recall: 0.99389, F1-Score: 0.99389
  17. Class-5 | Precision: 0.99549, Recall: 0.98991, F1-Score: 0.99269
  18. Class-6 | Precision: 0.99582, Recall: 0.99478, F1-Score: 0.99530
  19. Class-7 | Precision: 0.99033, Recall: 0.99611, F1-Score: 0.99321
  20. Class-8 | Precision: 0.99178, Recall: 0.99076, F1-Score: 0.99127
  21. Class-9 | Precision: 0.99105, Recall: 0.98811, F1-Score: 0.98958
  22. Total | Accuracy: 0.99370, Precision: 0.99371, Recall: 0.99364, F1-Score: 0.99367

Requirements

  • Python 3.7.6
  • Tensorflow 2.1.0
  • Numpy 1.18.1
  • Matplotlib 3.1.3

Reference

[1] Saining Xi et al. (2016). Aggregated Residual Transformations for Deep Neural Networks. arXiv preprint arXiv:1611.05431.