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
- 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:
-
Setup virtual environment
-
Install dependencies
pip3 install -r requirements.lock
-
Download the Standford Dogs dataset into
data/standford-dogs-dataset.zip -
Prepare dataset
python3 -m src.prepare_dataset [--images_per_race N]
-
Build similarity index
python3 -m src.build
-
Run similarity search
python3 -m src.run [--image_url https://example.org/image.jpg] [--image_path image.jpg] [--show]
Authors
License
MIT © Which dog are you?