solved1 / greyclassifier

CS_IOC5008 Visual Recognition using Deep Learning: Image Classification for grey natural scene images with <4000 annotated data

GreyClassifier

This repository gathers the code for greyscale natural image classification from the in-class Kaggle challenge.

Getting started

First, create a new virtual environment

virtualenv venv -p python3
source venv/bin/activate

You might need to make sure your python3 link is ready by typing

which python3

Then install the development requirements

pip install -r requirements.txt

Install pretrained weights

sh install_tools.sh

Training the base classifiers

Training configuration can be specified in src/configs.py. To train a model for a specific subclass, simply uncomment the desired SUBCLASS in this file and change LOGGER to rooms, nature or urban.

If you would like to train on single-channel images, you can set GREY = True.

Then, run:

python -m src.run

This will train the CNN model on the training and validation sets, then generate and save the concatenated outputs of the snapshot models in xgbdata.

Training the XGB meta-learners

Make sure that LOGGER in src/configs.py is set to the same one you used to train your base classifier, and that TRAIN = True

Run:

python -m src.ensemble

This will train and save the XGBoost model weights.

Ensemble prediction

First, set TRAIN = False in src/configs.py.

Run:

python -m src.ensemble

This will save the testing predictions under xgb.csv.

Future work:

• Add argument parsing so that the user does not have to edit the configuration file for each different run, and parameters can be passed as arguments instead