seedling-classification

My Capstone Project for Udacity's Machine Learning nanodegree will attempt to classify twelve different species of plants by image analysis. This project will utilize the dataset from Kaggle's Plant Seedlings Identification competition.

Setup

1. Download training images, test images. Rename train to labeled and test to unlabeled to distinguish them from the train/test split extracted from the labeled dataset.

2. Download Test Images from the Wild as described in this post.

Folder structure should look like this:

  seedling-classification
    - data
      - labeled
        - Black-grass
           - 0ace21089.png
             ...
        - Charlock
        ...
      - unlabeled
        - 0a64e3e6c.png
          ...
      - from_the_wild
        - Charlock
           - WP_20150506_10_15_19_Pro__highres_0.tiff
             ...
        - Cleavers
        ...

3. (Optional) If you plan to use GPUs, install the necessary NVIDIA software on your system. I followed this guide to set up GPUs on a Google Compute Engine instance.

4. Install anaconda if you don't have it already.

wget https://repo.continuum.io/archive/Anaconda3-5.0.1-Linux-x86_64.sh
bash Anaconda3-5.0.1-Linux-x86_64.sh
source ~/.bashrc

5. Create a new environment.

• Linux (to install with GPU support, change requirements/linux.yml to requirements/linux-gpu.yml):

conda env create -f requirements/linux.yml

• Mac (to install with GPU support, change requirements/mac.yml to requirements/mac-gpu.yml):

conda env create -f requirements/mac.yml

6. Create an IPython kernel for the seedlings environment.

python -m ipykernel install --user --name seedlings --display-name "seedlings"

7. Run this command to enable the tqdm_notebook extension

jupyter nbextension enable --py widgetsnbextension

8. Active the environment

source activate seedlings

9. Switch Keras backend to TensorFlow.

KERAS_BACKEND=tensorflow python -c "from keras import backend"

10. Open the notebook.

jupyter notebook --ip=0.0.0.0 --port=8888 --no-browser seedlings.ipynb

Acknowledgements

Setup Instructions were heavily influenced by Udacity's CNN dog-project