This repository is dedicated to training machine learning models using the QuickDraw dataset. It encompasses a range of utilities from classes for model training to functions designed for dataset setup. Moreover, the repository includes numerous notebooks that demonstrate the application of these classes and techniques.
One of the best models that has been fine-tuned is a Vision Transformer and has been pushed on hugging face 🤗: ilyesdjerfaf/vit-base-patch16-224-in21k-quickdraw
The QuickDraw-ML repository is crafted to support the development of machine learning models utilizing the QuickDraw dataset, a comprehensive collection of hand-drawn sketches gathered globally. This dataset offers a unique opportunity to train and refine models capable of recognizing and reproducing diverse sketches.
The repository contains several classes designed for the training of machine learning models on the QuickDraw dataset. These classes facilitate tasks such as data loading, preprocessing, and augmentation.
A suite of functions are provided to streamline the process of preparing the QuickDraw dataset. These include downloading, splitting into train/test sets, and preprocessing.
Below is the project structure, detailing the main folders and their contents:
| Folder/File | Description |
|---|---|
data/ |
Contains modules for generating and managing the dataset. |
mlruns/ |
Includes all experiments done. |
schema/ |
Contains all classes for training models in various ways. |
requirements.txt |
Lists all dependencies required by the project. |
To begin using QuickDraw-ML, follow these instructions:
- Clone the repository to your local machine.
git clone https://github.com/mlengineershub/QuickDraw-ML- Install required dependencies listed in
requirements.txt.
pip install -r requirements.txt- Run the dataset setup script to download and preprocess the QuickDraw dataset. (Ask someone for the
data/size_10000/folder)
python data/dataset_builder.py data/size_10000 data --max_images 250 --labels data/labels.txt