Lego bricks image classifier
- Create and activate virtualenv
alternatively, using virtualenvwrapper:
python3 -m virtualenv lego-ceptron source lego-ceptron/bin/activatemkvirtualenv --python=/usr/bin/python3 lego-ceptron
- Install packages
pip install -r requirements.txt
- Add project's root directory to PYTHONPATH:
add2virtualenv <project_root>
Download dalaset from this Kaggle
(login reqired) and unzip it to ./data directory.
Remember to unzip all subarchives as well, so that your file tree looks like this:
./data/
Base Images/
Cropped Images/
imageSetKey.csv
other .jpg files
After this go to ./data folder and run:
python split_data.py
Additional files train.csv, valid.csv and test.csv will be created.
From the project's root directory run:
python src/run.py train|predict 1|2|3|4
Allowed arguments:
$ python src/run.py --help
usage: run.py [-h] [-e EPOCHS] [-b BATCH_SIZE] [-ts {96,128,160,192,224}]
{train,predict} {1,2,3,4}
SNR Lego bricks classification 2019.
positional arguments:
{train,predict} Run mode.
{1,2,3,4} Task number: 1 - train FC layers only, 2 - train last
conv layer + FC, 3 - train whole model
optional arguments:
-h, --help show this help message and exit
-e EPOCHS, --epochs EPOCHS
Number of training epochs
-b BATCH_SIZE, --batch_size BATCH_SIZE
Batch size
-ts {96,128,160,192,224}, --target_size {96,128,160,192,224}
Target input image size.