fulhaus-furniture-classifier

A model that uses the MobileNet v2 architecture for classifying the Fulhaus furniture dataset.

Written in Python using Keras + Tensorflow, Served by FastAPI in Docker, CI/CD in Github Actions

Mean Validation Accuracy = 94.3% + 2.1% (33 validation files, n=20)

Workflow

0. unzip dataset
1. prepare-data.py (reorgnizes the original dataset for train-model.py)
2. train-model.py (feed the reorganized data directory into this script)

Accessing the model

The API file is located inside app/model/model.py

It takes a python list for its input where the first item in the list is a flattened (np.flatten) RGB image, and the second item in the list is the shape of the image (np.shape). A string label is returned for the prediction.

im = cv2.imread('A Chique Sofa.jpg') 
input = [im.flatten().tolist(), im.shape]
output = model.predict_pipeline(input)

Commands

Python

python prepare-dataset.py input output_dir
python train-model.py training_data

Docker Deployment

docker build -t furniture-detection-app .
docker run -p 80:80 furniture-detection-app

Heroku Deployment

heroku login
heroku create furniture-classifier-app-1