PyTorch Rocket Yolov3 RetinaNet50, RetinaNet101 - Tutorial 2: A Tale of 3 Rockets

Have you ever wanted to test multiple Deep Learning models and compare their results very easily?

Are you tired of picking a Deep Learning model just because it is the only one you are able to run?

We want to solve this problem and we packaged 3 state-of-the-art Deep Learning models for Object detection for you to easily test them.

We are calling those new way of packaging Deep Learning models: Rockets.

Welcome to the Rockets Scientists Community!!!

Install the repositories

We recommend you to use an isolated Python environement such as virtualenv or conda with at least Python 3.6. Then you can use the following lines of code:

git clone https://github.com/LucasVandroux/PyTorch-Rocket-YOLOv3-RetinaNet50-RetinaNet101
cd PyTorch-Rocket-YOLOv3-RetinaNet50-RetinaNet101
pip install rocketbase

Install PyTorch

As the installation for PyTorch is different for each platform, you need to look at the PyTorch installation guide. Don't worry it is very simple, maximum 2 lines of codes 😝

A Tale of 3 Rockets

For this first tutorial, we selected three state-of-the-art models in Object Detection for you to play with:

RetinaNet with a resnet50 backbone and smaller dimension resized to 608px [paper]
RetinaNet with a resnet101 backbone and smaller dimension resized to 800px [paper]
YOLOv3 [paper]

Note that the RetinaNet Rocket with the resnet50 backbone can be landed using igor/resnet. This Rocket is the default RetinaNet Rocket. To land another version we suggest to use a specific slug such as igor/retinanet-resnet101-800px.