Skip to content

CIFAR-10 trained with CreateML and a SwiftUI app for real time model testing. Kaggle Score: 0.69660

Notifications You must be signed in to change notification settings

josetorronteras/CIFAR10VisionClassifier

Repository files navigation

CIFAR10VISIONCLASSIFIER

Table of Contents

Introduction

This project was created to participate in the Kaggle competition for the CIFAR-10 dataset. The goal of the competition is to classify images from the dataset. Also a SwiftUI app was created to classify images from the camera or the photo library.

CIFAR-10 is a dataset that consists of several images divided into the following 10 classes:

  • Airplanes
  • Cars
  • Birds
  • Cats
  • Deer
  • Dogs
  • Frogs
  • Horses
  • Ships
  • Trucks

The dataset was trained using CreateML and the model was exported to a CoreML model. The model was then used in a SwiftUI app to classify images. This app can be used to classify images from the camera or the photo library.

CIFAR10VISIONCLASSIFIER

Training

For the training, the CIFAR-10 dataset was used. The dataset was downloaded from Kaggle. Parameters used for training:

  • Iterations: 100
  • Augmentation: Flip and rotate

training

SwiftUI App

The app was created using SwiftUI and the CoreML model was used to classify images. The app can be used to classify images from the camera or the photo library. Minimum iOS version: 15.0.


SwiftUI App Results

The app was tested using the following images (pictures taken by me):

Good results:


Bad results:


Installation

The installation is for testing purposes only.

python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
open CIFAR-10\ -\ Object\ Recognition\ in\ Images.ipynb

Kaggle Submission

kaggle Kaggle submission - Kaggle profile


Thanks.