Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/anatoliibobrov/number_recognition
Сравнение точности распознавания рукописных цифр с помощью двух видов нейронных сетей: MLP и CNN (PyTorch) | Comparing the accuracy of handwritten digit recognition using two types of neural networks: MLP and CNN (Pytorch)
https://github.com/anatoliibobrov/number_recognition
ai cnn mlp python3 pytorch
Last synced: 13 days ago
JSON representation
Сравнение точности распознавания рукописных цифр с помощью двух видов нейронных сетей: MLP и CNN (PyTorch) | Comparing the accuracy of handwritten digit recognition using two types of neural networks: MLP and CNN (Pytorch)
- Host: GitHub
- URL: https://github.com/anatoliibobrov/number_recognition
- Owner: AnatoliiBobrov
- Created: 2024-09-01T15:00:45.000Z (2 months ago)
- Default Branch: master
- Last Pushed: 2024-09-07T09:15:28.000Z (2 months ago)
- Last Synced: 2024-10-06T02:06:15.405Z (about 1 month ago)
- Topics: ai, cnn, mlp, python3, pytorch
- Language: Python
- Homepage:
- Size: 10.7 MB
- Stars: 0
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
The English translation is below.
В этом репозитории мы сравним точность распознавания тестового набора цифр MNIST.
Для это мы описали для модели нейронной сети: MLP (Net) и CNN + MLP (ConvNet).
Набор цифр в файле "data.bin" имеет вид: (x_train, y_train), (x_test, y_test).
Обучение нейронных сетей производится с помощью модуля PyTorchРеализация структуры второй модели взаимствована [отсюда](https://www.educba.com/pytorch-conv2d/).
Сверточная нейронная сеть дает точность на 1% большую, чем простой многослойный
персептрон, уступая во времени обучения в 6 раз.Результат выполнения представлен ниже (время в секундах)
English
In this repository, we will compare the recognition accuracy of the MNIST test set of digits. To do this, we have described for the neural network model: MLP (Net) and CNN + MLP (Convention). The set of digits in the "data.bin" file has the form: (x_train, y_train), (x_test, y_test). Neural networks are trained using the PyTorch moduleThe implementation of the structure of the second model is borrowed from [here](https://www.educba.com/pytorch-conv2d/). A convolutional neural network provides 1% greater accuracy than a simple multilayer perceptron, yielding 6 times in training time.
The result of the execution is shown below (time in seconds)
```
Training of MLP...
...
Epoch: 9, accuracy: 99.51%, time: 110.72
Survey accuracy: 98.11%, time: 2.08
Overall time: 1131.75
Training of CNN + MLP...
...
Epoch: 9, accuracy: 99.34%, time: 637.07
Survey accuracy: 99.26%, time: 28.40
Overall time: 6604.93
```