aspctu/autologo

generating synthetic training data for logo detection in images

AutoLogo

Automatically generate object detection networks via one logo image

Usage

1. run gen_data.py with following arguments:

   python gen_data.py --logo (path to logo image) --label (class name)

2. The above command will push 2 files into the csvFiles directory:

   [label].csv : Holds the CSVGenerator file needed to train the network in the next step.

   label.csv : Attaches label to key value, needed to train as well

3. The new synthetic dataset will be placed in /syntheticImages

4. CD into the retinanet/keras_retinanet/bin directory

5. Run train.py with the following arguments:

   python train.py <path/to/csvFiles/[label].csv> <path/to/csvFiles/label.csv>

   Trianing will begin and snapshots will be saved to the snapshot folder

Acknowledgements

Creds to @fizyr for the keras-retinanet model definition.