Training¶
There are two types of training currently supported: classifier and detection.
Structure¶
Most machine learning model training pipelines can be broken down into a few major components: the model, dataset, and loss functions.
Examples¶
Classification:
PYTHONPATH=. hawk_eye/train/train_clf.py \
--config hawk_eye/configs/vovnet.yaml
Detection:
PYTHONPATH=. hawk_eye/train/train_det.py \
--config hawk_eye/configs/vovnet-det.yaml
Fine Tuning¶
It is an unfortunate fact that most deep learning models need to be trained on GPUs to be done in a reasonable amount of time.
If you only have a CPU, you might want to finetune a model. This means taking an model intially trained on lots of data and training it on yours. This can drastically speed up training.