Training a model for object-detection is quite a bit more convoluted than just classification. But the advantage is knowing exactly where and what is in the image.
The process for training a model for object-detection is described in the file:
training-object-detection
Todo..
I made a fairly good guide a while ago in another repository.
Just one one of these Docker containers instead:
Using CPU
docker run --rm -it -p 8001:8888 -p 6006:6006 -v $HOME/andet/training/docker-training-shared/object-training:/root/sharedfolder:Z jacobpeddk/tensorflow-improved:latest
Using GPU
nvidia-docker run --rm -it -p 8001:8888 -p 6006:6006 -v $HOME/andet/training/docker-training-shared/object-training:/root/sharedfolder:Z jacobpeddk/tensorflow-improved:latest-gpu