README.md

Easy Image Classification Model Maker

Create a custom image classification model with a few lines of code. This module scrapes images, formats and uploads the image dataset to 🤗, and trains a 🤗 model. Built on top of 🤗 Transformers and 🤗 Datasets.

Installation

pip install -r requirements.txt

Train Model

Import the module

import modelmaker

Define the model and dataset parameters:

• keyword list of strings will be the labels of the model
• num_images number of images in the training dataset
• key HuggingFace write access token can be created here.
• dataset_name name of dataset that will uploaded to HuggingFace
• model_name name of model that will be uploaded to HuggingFace
• train_epochs number of training epochs the model will go through

model = modelmaker.ModelMaker(keywords = ['cubism', 'impressionism', 'abstract expressionism'],
                              num_images = 100,
                              key = 'YOUR_TOKEN',
                              dataset_name = 'art_dataset',
                              model_name = 'art_classifier',
                              train_epochs = 10)

Download images from Bing into the './images' folder. It is suggested to manually go through the image folders to make sure there isn't any incorrect images in their respective folders.

model.download_images()

Upload dataset to HuggingFace

model.upload_dataset()

Train the model and upload it to HuggingFace

model.train_model()

Model Usage

Inference API Widget

Go to the model page, which can be found on your HuggingFace page. Drag and drag images onto the Inference API section to test it.

Python

from transformers import pipeline

pipe = pipeline("image-classification", model="tonyassi/art_classifier")
result = pipe('image.png')

print(result)

JavaScript API

async function query(filename) {
	const data = fs.readFileSync(filename);
	const response = await fetch(
		"https://api-inference.huggingface.co/models/tonyassi/art_classifier",
		{
			headers: { Authorization: "Bearer xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx" },
			method: "POST",
			body: data,
		}
	);
	const result = await response.json();
	return result;
}

query("art.jpg").then((response) => {
	console.log(JSON.stringify(response));
});