Skip to content

Latest commit

 

History

History
74 lines (55 loc) · 1.83 KB

README.md

File metadata and controls

74 lines (55 loc) · 1.83 KB
Raw

AI-Generated-Zoology

Our goal is to be able to convert an animal sketch to an animal picture.

Set-up project

  1. Git clone the repository
git clone https://github.com/niyonx/AI-Generated-Zoology.git
  1. Create your python virtual environment
virtualenv --python=/usr/bin/python3.8 venv
source venv/bin/activate
  1. Install dependencies
pip install -r requirements.txt

Run Jupyter Notebooks:

Locally: Make sure jupyter is using venv python interpreter.

jupyter lab

Images to sketch

If you start from scratch, then you only have images with background removed by some means. We used photoshop.

  1. Place all your images in
data/raw-img/
  1. Edge detection model only works on jpeg files, so convert all images by running
./scripts/png_to_jpeg.py
  1. From PhotoSketch directory, run
./scripts/test_pretrained.sh (Linux)

./scripts/test_pretrained.cmd (Windows, can run from any directory or just double click on file)
  1. Sketch images should be located in:
./data/sketh-img/

Project data

For demo purposes, we placed only a small portion of the datasets we used.
The full data we used for the project can be downloaded from: https://drive.google.com/file/d/1pzdkQ1OBVexu6YrW_ULHg_lUDiOu2ef1/view?usp=sharing

Checkpoints

We saved our models after 500 epochs for both cats and dogs datasets.
However, due to their large size, we make them available at: https://drive.google.com/file/d/1TH6DImrtBW70vWCeLD5ro7tA_LRlv4qG/view?usp=sharing

Demo

Simply run all the cells up to the RESULTS section.
DO NOT run the cells below RESULTS.
We ran those cells with the checkpoints loaded to showcase our results.
You would need to download the checkpoints to run those cells.

We have added some of our results in data/results(dogs) and data/results(cats)