/chestnut-seaurchin-classifier

Classify sea urchins and chestnuts with machine learning

Primary LanguagePythonMIT LicenseMIT

chestnut-seaurchin-classifier

Our goal

Our goal is to accurately classify sea urchins and chestnuts. The ultimate goal is to correctly predict chestnuts in the water.

Article (Japanese)

https://qiita.com/baibai25/items/93dfd703ccba6e097ba5

Demo

You can easily try to demo.

  1. Download model https://drive.google.com/open?id=1Yoq3Nwx72zXTtANvEqp_aNCvoo3wtk_A

  2. Run with YOLO

./darknet detector test cfg/foo.data cfg/foo.cfg unikuri_best.weights test_img.jpg

Translation

kuri means chestnut. uni means sea urchin.

Scraping

-k: Search words
-l: Number of images to download

python download.py -k keywords A, keywords B -l 1000

Data augmentation

-i: Input directory
-o: Output directory

python augmentation.py -i input_dir -o output_dir

Building a object detector using YOLO

  1. Download

    git clone https://github.com/AlexeyAB/darknet.git
cd darknet/
make

wget https://pjreddie.com/media/files/darknet53.conv.74

  2. Move labeled data

  3. Define learning settings
    cfg/foo.data
    cfg/foo.names
    cfg/foo.cfg

  4. Training

    ./darknet detector train cfg/foo.data cfg/foo.cfg darknet53.conv.74

For more details, refer to the https://github.com/AlexeyAB/darknet?files=1.

Result