Train

Now that you’ve got your data in a format the model expects, let’s train an EfficientDet to do object detection on our dataset.

Within the root directory of the project, run the following command

python trashnet/train.py --num_epochs NUM_EPOCHS --path DATA_PATH/dataset

where NUM_EPOCHS is the number of epochs you want to train the network for and DATA_PATH is the path to the COCO style dataset that you set up previously.

For additional hyperparameter choices available during training, use help

python trashnet/train.py --help

The training loop comes with default hyperparameters that have been tested to work on the dataset, but feel free to try and experiment.

Depending on the underlying hardware and the number of epochs you’re training for, it can take anywhere from a couple of minutes to a day for the network to finish training. Go ahead and grab a coffee while the network learns.