This directory contains a collection of Jupyter Notebooks and Scripts that can be used to build datasets and train a model. The scripts are designed to automate most of the steps in the process. If you want to walk through the process yourself, checkout the notebooks.

Since you will be training a model, an Nvidia GPU is required. In order for a Docker container to access the GPU, the Nvidia Container Toolkit needs to be installed. There are directions for doing that here. You can test to make sure everything is installed correctly and working, with the following command:

sudo docker run --rm --gpus all nvidia/cuda:11.0-base nvidia-smi

In the main directory of this repository, run the following command:

docker-compose build

In the main directory of this repository, use Docker Compose to launch the container:

docker-compose up

The following directories will be created when the container is launched for the first time. They are mapped into the container under /tf/. Here is what they are used for:

• model-export Trained models are exported here
• dataset-export When a dataset is exported
• notebooks This is where the wonderful notebooks are stored, this way if you make changes to them they serve the container restarting
• testing This is for passing in raw images that were captured
• fiftyone-db This gets mapped to the location where the Voxel51 DB for the datasets is saved. This lets them stay around between container restarts.
• models The TF Model repo gets installed here.

To run the automated scripts, you will want to attach to a shell inside the container:

sudo docker exec -it ml-model_jupyter_1  /bin/bash
cd scripts

Now you are ready to start running scripts. Check out the documentation here.

Goto to the IP/domain name for the computer this is running on.... probably localhost. and port 8888 in a browser: https://localhost:8888

This will bring up the list of folders in the Jupyter client, go into the notebooks folder.

Here is what the following notebooks help you do and the rough order you want to do them in:

*We use LabelBox to label our images. The free tier should support most tasks. Prior to starting, create a Labelbox Project and associated dataset. You will need the Labelbox project & dataset name, as well as a Labelbox API.

• Create Voxel51 Dataset.ipynb Run this first to load images from the testing directory into a Voxel51 dataset
• Add FAA Data to Voxel51 Dataset.ipynb Adds labels to the Voxel51 Dataset based on the FAA data
• Upload from Voxel51 to Labelbox.ipynb Sends the samples to Labelbox for annotation

After you have finished uploading the images, go to Labelbox and label the images. Then export the labels as a JSON file. Download the JSON file and move it to the machine that is running the Docker container that is serving the Notebook.

• Export from Labelbox to Voxel51.ipynb Export the annotations from Labelbox
• Train Plane Classification Model.ipynb
• Add Plane Classification to Voxel51 Dataset.ipynb

• Export from Labelbox to Voxel51.ipynb Export the annotations from Labelbox

• Train Plane Detection Model.ipynb

• Add Object Detection to Voxel51.ipynb

• Examine TFRecord.ipynb You can use this if you want to check out what is in a TFRecord

If you wish to access the running version of the container and poke around on the command line inside it, use the following command:

sudo docker exec -it ml-model_jupyter_1  /bin/bash

If you wish to monitor the progress of a model being trained, Tensorboard provides a nice visualization of different metrics. To launch it, use the command above to attach to the running container and then run the following:

tensorboard --logdir=/tf/training/ --bind_all

If you goto port 6006 of the machine where the container is running in a browser, the Tensorboard app should pop up.

python object_detection/model_main_tf2.py \
    --pipeline_config_path=/tf/models/research/deploy/pipeline_file.config  \
    --model_dir=/tf/training/d0_plane_detect \
    --checkpoint_dir=/tf/training/d0_plane_detect \
    --alsologtostderr