To download the full dataset, install Resilio Sync and sync this folder (27.6 GB).
If you would like to download a sample of the dataset to see if it matches your needs, here's a few sample sets:
- Small sample set, 200 images each class. Total zipped size: 707 MB
- Large sample set, 500 images each class. Total zipped size: 1.56 GB
Each image is 1164x874.
To start training, first unzip one of the datasets into the root directory of traffic-lights so that the directory tree looks like: /traffic-lights/data/GREEN/etc.png.
When you run train.py, it will then start a process of cropping and randomly transforming images from the dataset into the /traffic-lights/data/.processed directory.
Depending on the amount of data and your CPU, you may want to decrease the amount of data generator threads as defined here as it's pretty heavy on system resources. An ETA will print approximately every 15 seconds to give you feedback on how long it will take.