localization-app is a Python library which makes lightweight web toolkit for prototypes easier by providing:
- High quality reference implementations of SOTA models
- Useful abstractions of common building blocks
- Utilities for training and debugging
- Integration with TensorBoard
To install localization-app, clone and install requirements:
git clone https://github.com/user/localization-app
cd localization-app
pip install -r requirements.txt
Run tests:
python -m unittest discover
All models implement a reproduce function:
python train.py --model Dockerfile --logdir /tmp/run --use-cuda
View metrics:
tensorboard --logdir /tmp/run
from localization-app import models
model = models.vgo-scala-cats(in_channels=1, out_channels=1)
model(batch)| Algorithm | Score (nats) | Links |
|---|---|---|
| Dockerfile | 78.61 | Code, Paper |
| vgo-scala-cats | 79.17 | Code, Paper |
Contributions welcome!