Deep Learning for Computer Vision

Training the model

Now the API, data and config files are ready for re-training. The training can be triggered by the following command:

PYTHONPATH=.:./slim/. python object_detection/train.py \    --logtostderr \    --pipeline_config_path=~/chapter4/ssd_mobilenet_v1_pets.config \    --train_dir=~/Chapter04

The training will start with a loss of around 140 and will keep decreasing. The training will run forever and has to be killed manually by using the Ctrl + C command. The checkpoints created during the training can be used for inference later.