https://github.com/yeonghyeon/rexnet-tf2
TensorFlow implementation of "ReXNet: Diminishing Representational Bottleneck on Convolutional Neural Network"
https://github.com/yeonghyeon/rexnet-tf2
convolutional-neural-network convolutional-neural-networks mnist mnist-classification mnist-dataset state-of-the-art
Last synced: 5 months ago
JSON representation
TensorFlow implementation of "ReXNet: Diminishing Representational Bottleneck on Convolutional Neural Network"
- Host: GitHub
- URL: https://github.com/yeonghyeon/rexnet-tf2
- Owner: YeongHyeon
- License: mit
- Created: 2020-07-03T08:32:30.000Z (over 5 years ago)
- Default Branch: master
- Last Pushed: 2020-07-05T23:43:07.000Z (over 5 years ago)
- Last Synced: 2023-10-20T23:55:49.239Z (almost 2 years ago)
- Topics: convolutional-neural-network, convolutional-neural-networks, mnist, mnist-classification, mnist-dataset, state-of-the-art
- Language: Python
- Homepage:
- Size: 266 KB
- Stars: 9
- Watchers: 4
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
[TensorFlow 2] ReXNet: Diminishing Representational Bottleneck on Convolutional Neural Network
=====
Unofficial TensorFlow implementation of "ReXNet: Diminishing Representational Bottleneck on Convolutional Neural Network".
Official PyTorch implementation is provided as the following link.
https://github.com/clovaai/rexnet
* Note that, the detailed neural network architecture is different with the original paper [1].
* This repository only contains the shorten neural network reflecting the concept.
## Concept
Spec of ReXNet-v1 [1].
## Performance
|Indicator|Value|
|:---|:---:|
|Accuracy|0.98180|
|Precision|0.98188|
|Recall|0.98177|
|F1-Score|0.98171|
```
Confusion Matrix
[[ 974 0 0 0 0 0 4 1 0 1]
[ 2 1120 1 1 1 1 2 2 3 2]
[ 6 0 1008 0 4 0 0 4 7 3]
[ 2 0 5 968 0 10 0 3 6 16]
[ 1 0 0 0 964 0 4 0 0 13]
[ 2 0 0 4 1 874 3 1 0 7]
[ 6 2 0 0 1 4 943 0 1 1]
[ 0 1 4 1 2 0 0 1010 1 9]
[ 7 0 2 0 0 2 0 1 954 8]
[ 0 2 0 1 1 2 0 0 0 1003]]
Class-0 | Precision: 0.97400, Recall: 0.99388, F1-Score: 0.98384
Class-1 | Precision: 0.99556, Recall: 0.98678, F1-Score: 0.99115
Class-2 | Precision: 0.98824, Recall: 0.97674, F1-Score: 0.98246
Class-3 | Precision: 0.99282, Recall: 0.95842, F1-Score: 0.97531
Class-4 | Precision: 0.98973, Recall: 0.98167, F1-Score: 0.98569
Class-5 | Precision: 0.97872, Recall: 0.97982, F1-Score: 0.97927
Class-6 | Precision: 0.98640, Recall: 0.98434, F1-Score: 0.98537
Class-7 | Precision: 0.98826, Recall: 0.98249, F1-Score: 0.98537
Class-8 | Precision: 0.98148, Recall: 0.97947, F1-Score: 0.98047
Class-9 | Precision: 0.94356, Recall: 0.99405, F1-Score: 0.96815
Total | Accuracy: 0.98180, Precision: 0.98188, Recall: 0.98177, F1-Score: 0.98171
```
## Requirements
* Python 3.7.6
* Tensorflow 2.1.0
* Numpy 1.18.1
* Matplotlib 3.1.3
## Reference
[1] Dongyoon Han et al. (2020). ReXNet: Diminishing Representational Bottleneck on Convolutional Neural Network. arXiv preprint arXiv:2007.00992.