[TensorFlow 2] Aggregated Residual Transformations for Deep Neural Networks

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

Related Repositories

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

Concept

The three ways for construct ResNeXt block [1].

Performance

Confusion Matrix
[[ 977    0    1    0    0    0    0    0    2    0]
 [   0 1129    2    0    0    0    1    2    1    0]
 [   0    1 1026    0    1    0    0    3    1    0]
 [   0    0    2 1007    0    1    0    0    0    0]
 [   0    0    0    0  976    0    1    0    0    5]
 [   1    0    0    4    0  883    2    0    0    2]
 [   1    1    0    0    1    1  953    0    1    0]
 [   0    1    2    0    0    0    0 1024    1    0]
 [   2    0    2    1    0    0    0    2  965    2]
 [   0    1    0    0    4    2    0    3    2  997]]
Class-0 | Precision: 0.99592, Recall: 0.99694, F1-Score: 0.99643
Class-1 | Precision: 0.99647, Recall: 0.99471, F1-Score: 0.99559
Class-2 | Precision: 0.99130, Recall: 0.99419, F1-Score: 0.99274
Class-3 | Precision: 0.99506, Recall: 0.99703, F1-Score: 0.99604
Class-4 | Precision: 0.99389, Recall: 0.99389, F1-Score: 0.99389
Class-5 | Precision: 0.99549, Recall: 0.98991, F1-Score: 0.99269
Class-6 | Precision: 0.99582, Recall: 0.99478, F1-Score: 0.99530
Class-7 | Precision: 0.99033, Recall: 0.99611, F1-Score: 0.99321
Class-8 | Precision: 0.99178, Recall: 0.99076, F1-Score: 0.99127
Class-9 | Precision: 0.99105, Recall: 0.98811, F1-Score: 0.98958
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.