PyTorch Implementation of PlaceNet

Dependencies

(optional)

Train: run.sh
- This script basically uses all GPUs installed in the computer.
- To use single GPU,
  - remove ddp, and
  - add gpu=<the number of gpu you want to use>

You can also use your own 3D scenes to train PlaceNet.
It will be helpful to use the House-Traveler, a trajectory generator that create natural paths in 3D environments.
https://github.com/Yoo-Youngjae/house_traveler

To implement a 3D scene representation and rendering approach, we referenced the Generative Query Networks (GQN).
In the course of developing this project, we referenced various GQN implementations:
- https://github.com/ogroth/tf-gqn
- https://github.com/wohlert/generative-query-network-pytorch
- https://github.com/masa-su/pixyzoo/tree/master/GQN
- https://github.com/iShohei220/torch-gqn
- Among them, I mostly refer to iShohei220's work, which is composed of the most intuitive codes.