Library of different object detection algorithms. You need a CUDA compatible GPU for this to work.

This project uses trunk for linting and formatting. Check https://trunk.io/products/check

For dependency, poetry is used: https://python-poetry.org/

Go to root directory and run poetry install. But first make sure you have CUDA in your system along with all necessary packages. You can follow https://github.com/proafxin/cuda-gpu-scripts for CUDA related instructions. Make sure CUDA can interact with your GPU. The best way to ensure that is to go to python shell and run

import torch
torch.cuda.is_available()

You should see True as output. Finally, for virtual environment management, automated testing and task management, tox is used: https://tox.wiki/

Install tox in your system using python3 -m pip install -U tox and then run tox from the project directory.

The idea behind the structure in this package is that we want to make the model reusable and easily configurable. We will say, a model consists of some units. Each such unit can consist of layers or blocks or cells. Cell is an individual component such as Convolution, MaxPool, ReLU or other activation function, etc. A block consists of some cells. Sometimes a single component is repeated multiple times. To make it easy for us to specify the model, instead of adding the repeated components individually, we simply specify a repeition factor. The same idea goes for a layer. A layer consists of some blocks which may be repeated. The idea of unit may be confusing at first. First of all, it completely depends on the user. However, a good idea to make a separation would be at where we take maxpool or connected layers. See the following figure: Here, there is a maxpool after the first convolution. So that will be our first unit: convolution combined with maxpool. Same for second unit. In third unit, we have 4 convolution and then maxpool. In fourth unit, we have two convolutions which are repeated, then two convolutions and a maxpool layer, and so on.