REVA-QCAV/README.md

sphereDetect

sphereDetect is a simple neural network, based on a Mask R-CNN, to detect spherical landmarks for image calibration.

Built with

• Python

Frameworks

• PyTorch
• TorchVision
• PyTorch Lightning
• PyTorch Lightning Bolts
• PyTorch Metrics
• ONNXRuntime

Tools

• Poetry
• Docker
• VSCode
  • ms-python
  • Python Docstring Generator
  • Conventional Commits
  • Remote container
  • EditorConfig
  • Docker
  • Jupyter

Getting started (with docker and vscode)

Requirements

• Git
• Docker
• NVIDIA-Docker

Installation

Clone the repository:

git clone git@git.inpt.fr:fainsil/pytorch-reva.git

Start VS Code:

vscode pytorch-reva

Install the Remote Development extension pack.
Modify variables UID and GID in .devcontainer/Dockerfile if necessary. Reopen the workspace in devcontainer mode.

Usage

Configure Weights & Biases (local) server at http://localhost:8080, and login:

wandb login --host http://localhost:8080

Press F5 to launch src/train.py in debug mode (with breakpoints, slower)
or press Ctrl+F5 to launch src/train.py in release mode.

Getting started (without docker)

Requirements

• Git
• Poetry
• Python
• Docker (if local wandb server used)

Installation

Clone the repository:

git clone git@git.inpt.fr:fainsil/pytorch-reva.git
cd pytorch-reva

Install the dependencies:

poetry install --with all

Usage

Activate python environment:

poetry shell

Configure Weights & Biases (local) server, and login:

wandb server start
wandb login --host http://localhost:8080

Start a training:

python src/train.py

License

Distributed under the MIT license.
See LICENSE for more information.

Contact

Laurent Fainsin [loʁɑ̃ fɛ̃zɛ̃]
<laurent@fainsin.bzh>