kayoyin / digit-detector

Retinanet Digit Detector

This repository contains the code to train a Retinanet digit detector from scratch using the SVHN dataset.

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

Data pre-processing

Download the training and testing data, extract and put it in your project as below:

|-- dataset
	|-- train
	|-- test

The project uses the Pascal VOC format for the training annotations. To process the annotations, run:

sh preprocess.sh

You should obtain

|-- dataset
	|-- train_anns
        |-- train_imgs
	|-- test

Training

Before training the model, split your training set into a train and a validation set as below.

|-- dataset
	|-- train_anns
        |-- train_imgs
        |-- val_anns
        |-- val_imgs
	|-- test

Then, run the following to train the model.

python retinanet/main.py

Evaluation

Running the following will generate the predictions of the model for the test set and save them under submission.json

python retinanet/predict.py