Which dog are you?

🐶 Dog and face person matcher based on image similarity

Results

Input

Output

Summary

Inspiration

This project came to my mind in order to improve the experience seen in all the Instagram filters where everything is random. With that being said, I wanted to really know which dog is the one that is most similar to me. Yeah, it's random. I know.

What it does

From an image (image path or image URL), it returns the most similar dog given the Standford Dogs dataset. Simple and beautiful.

How I built it

Using Python 3.7, Tensorflow, NMSLIB, NumPy and Pillow, it builds a similarity index where I extract the features of the input image versus the built index and get the closest one.

Challenges I ran into

To be honest, I wanted to do another project (completely different from this one) in this hackathon (HackFromHome Round 2) but I got stuck after several hours and realized that was impossible. Yeah, what a pity.

That's why I decided to move to this project.

Accomplishments that I'm proud of

Building this random thing in less than 3 hours.

What's next for Which dog are you?

A lot of things, like a beautiful UI and getting this deployed somewhere.

Project

Requirements

1. Python 3.7+.

Recommendations

Usage of virtualenv is recommended for package library / runtime isolation.

Usage

To run the pipeline, please execute the following from the root directory:

1. Setup virtual environment

2. Install dependencies

   pip3 install -r requirements.lock

3. Download the Standford Dogs dataset into data/standford-dogs-dataset.zip

4. Prepare dataset

   python3 -m src.prepare_dataset [--images_per_race N]

5. Build similarity index

   python3 -m src.build

6. Run similarity search

   python3 -m src.run [--image_url https://example.org/image.jpg] [--image_path image.jpg] [--show]

Authors

• Albert Suàrez

License

MIT © Which dog are you?