A minimal template for training an object detection model

This project demonstrates transfer learning using Tensorflow. The underlying concepts provide a solid foundation for training machine learning models.

Quick Start: Setup the development environment

1. Install Anaconda or Miniconda.

2. Create a virtual environment with Python 3.7.3, no default packages, and activate it.

   conda create --name myenv --no-default-packages python=3.7
   conda activate myenv

3. Install the required packages.

   # from the repository's root directory:
   pip install -r requirements.txt

4. Install ipykernel:

   python -m ipykernel install --user --name=myenv

Quick Start: Collect and label images

1. Collect target images in ./data/raw/.

2. Label images using labelImg.

   a. From terminal run:

   labelImg

   b. Using labelImg, navigate to ./data/raw/ and label images.

Running Project: Using CLI

1. Setup:

   python cli.py --setup --image-format < > --pretrained-model-url < > --pretrained-model-name < >

   E.g.

   python cli.py --setup --image-format jpg --pretrained-model-url http://download.tensorflow.org/models/object_detection/tf2/20200711/ssd_mobilenet_v2_fpnlite_320x320_coco17_tpu-8.tar.gz --pretrained-model-name ssd_mobilenet_v2_fpnlite_320x320_coco17_tpu-8

2. Train:

   python cli.py --train --batch-size < > --epochs < > --train-test-split < > --pretrained-model-ckpt < >

   E.g.

   python cli.py --train --batch-size 4 --epochs 1000 --train-test-split 0.7 --pretrained-model-ckpt 0

3. Smoke test:

   python cli.py --validate --trained-model-ckpt < >

   E.g.

   python cli.py --validate --trained-model-ckpt 1

Running Project: Using notebooks

From the repository's root directory, start a Jupyter Notebook session and select the kernel initialized during setup.

1. Step through ./notebooks/setup.ipynb to set up the project.

2. Step through ./notebooks/train.ipynb to train an object detection model.

3. Step through ./notebooks/validate.ipynb to smoke test the trained image-object detection model.

---

N.b., to enable GPU-base training, check your TensorFlow version and ensure tallying CUDA and CUDNN versions are installed.

Promotion

👏 Star if you liked

👐 Share if you loved

🎓 Fork to contribute