https://github.com/javadr/pytorch-cifar10

Using PyTorch to predict CIFAR10 images' labels with 88% accuracy.
https://github.com/javadr/pytorch-cifar10

cifar10 classification pytorch pytorch-cnn

Last synced: 4 days ago
JSON representation

Using PyTorch to predict CIFAR10 images' labels with 88% accuracy.

• Host: GitHub
• URL: https://github.com/javadr/pytorch-cifar10
• Owner: javadr
• License: gpl-3.0
• Created: 2021-09-20T18:46:29.000Z (over 3 years ago)
• Default Branch: main
• Last Pushed: 2022-03-06T19:42:19.000Z (almost 3 years ago)
• Last Synced: 2024-11-14T20:46:19.628Z (2 months ago)
• Topics: cifar10, classification, pytorch, pytorch-cnn
• Language: Jupyter Notebook
• Homepage:
• Size: 50.5 MB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# PyTorch-CIFAR10

Using PyTorch, the following project implements a deep neural network for predicting the class of input images based on [CIFAR10](https://www.cs.toronto.edu/~kriz/cifar.html) dataset. The dataset has 10 classes including: ‘airplane’, ‘automobile’, ‘bird’, ‘cat’, ‘deer’, ‘dog’, ‘frog’, ‘horse’, ‘ship’, ‘truck’. All images are 3-channel color images of 32x32 pixels. The image below, taken from [PyTorch website](https://pytorch.org/tutorials/beginner/blitz/cifar10_tutorial.html), shows 10 sample images from each class. 



Final accuracy is around 88%. 

### Model Architecture (Deep Net)

![Model Architecture](images/cifar10-arch.png)

#### Model Accuracy and Loss

Model Accuray | Model Loss 

:--------:|:-------:

|

#### Confusion Matrix 



Although the most misclassified classes are `bird` and `cat`, as it is obvious, the most confused ones with each other are `cat` and `dog` images. 

### Misclassification Samples



### Requirements

* [PyTorch](https://github.com/pytorch/pytorch)

* [torchvision](https://github.com/pytorch/vision)

### TODO

1. [ ] Train the model with Data Augmentation

2. [ ] Applying the Half Precision mechanism

3. [ ] Redesigning the Model  to get better results

4. [X] Showing some images from misclassification

#### Contents

1. `Cifar10.ipynb` - IPython notebook which implements the model 

2. `draw_convnet.py` - Code for drawing the model architecture taken from [draw_convnet](https://github.com/gwding/draw_convnet)

3. `images` folder - Contains the pictures of `Accuracy`, `Loss`, and `Confusion Matrix` in addition to `Model Architecture`. 

4. `cifar10.h5` - Trained model with accuracy of 88%.