Detectron2 Tutorial

A brief introductory tutorial to the Detectron2 library.

For more information, refer to the blog post

By Sebastian Castro, 2020

---

Getting Started

First, you will need a Linux or macOS environment set up with Python 3.6 or higher.

• Create a new Python virtual environment, e.g., python3 -m venv .virtualenvs/detectron2

• Activate the virtual environment, e.g., source .virtualenvs/detectron2/bin/activate

• pip install wheel

• pip install jupyter torch torchvision

• Install Detectron2 as outlined in the Detectron2 install guide. For Torch 1.7 / CUDA 11.0 (which is what was used for developing this tutorial), the command is:

python -m pip install detectron2 -f \
  https://dl.fbaipublicfiles.com/detectron2/wheels/cu110/torch1.7/index.html

• Set up OpenCV. Detectron2 requires Python 3.6 or higher so you will have to install from source according to this resource: https://docs.opencv.org/master/d2/de6/tutorial_py_setup_in_ubuntu.html

---

Labeling Data

We have used coco-annotator to manually label our data and export it to the popular COCO data format.

This repository includes a small dataset in the data folder so you can try Detectron2 without having to collect and label your own data.

---

Using Detectron2 for Object Detection

Once your environment is set up, you can follow along the notebooks in this repository.

• detectron_pretrained.ipynb shows how to use pretrained models from the Detectron2 Model Zoo.
• detectron_tutorial.ipynb shows how to train a model on a custom dataset by starting from one of the pretrained models above.