Intel Image Classifcation with CNN, Keras

The main goal of the project is to build a Tensorflow Convolution Neural Network (CNN) model that can classify images correctly and reliably.

In this work, I will try to come up with a reliable manually-tuned CNN model then compare it with pre-trained models (MobileNet, VGG16, ResNet16). We'll apply different techniques to increase model performance and reliability.

Dataset: Intel Image Classification

"This is image data of Natural Scenes around the world." - Based on the author

There are about 25k images of size 150x150 in 6 categories:

• buildings -> 0
• forest -> 1
• glacier -> 2
• mountain -> 3
• sea -> 4
• street -> 5

The data contains Train, Test and Prediction sets, each is compressed in a single zip. The ratio between three sets is 6:1:3, with 14k images in the train set, 3k in the test set and 7k in prediction set.

This data was initially published on a datahack by Intel for a Image classification Challenge.

Project goals

• Develop machine learning (ML) models with convolutional neural network (CNN) technology.
• Build a ML model that can classify the images reliably with good/reliable accuracy.
• Training different CNN models including self-built and pre-trained to find the most suitable-efficient model for the task.

Working process

1. Explore the images and prepare the datasets for training. We will process the images into Tensorflow datasets for training/testing/predicting.
2. CNN model training & prediction:
   • Build & train a CNN model.
   • Feature extraction with different pre-trained models.
   • Apply hyperparameter tuning with keras_tuner
   • Conduct model comparison, error analysis.

How to start

• Install requirements and dependencies in readme.md and Makefile.
• Download dataset from Kaggle and unzip it to data/raw, there should be 3 folders: seg_train, seg_pred and seg_test.
• Processed data should be put on data/processed.
• File can be downloaded directly using Kaggle API but would require additional setups

Repository tree

├─ docs/ - Project documents
│  ├─ figures/
│  ├─ other files: explanation to some steps and the jupyer notebook in pdf/html
├─ models/ - Saved TF models used in the project
├─ src/ - Source code
├─ Makefile
├─ intel_img_classification.ipynb: Jupyter notebook of the project
├─ project-description.ipynb: Full project description and dashboard local run tutorial
├─ README.md
├─ requirements.txt
├─ pyproject.toml
└─ setup.cfg