Neural Poisson Surface Reconstruction from Unoriented Point Clouds

This project is the project of my thesis about surface reconstruction.

Installation

TODO: Write here how we can install it.

Ablation (Dynamic Folder)

Example

The base way to run ablations when using a config file:

python train.py +ablation=learning_rate

If you want to use the CLI:

python train.py +ablation=base task_name=learning_rate model.optimizer.lr='choice(1e-03,1e-04)'

Debug

python train.py +ablation=learning_rate +debug=ablation

Sweep (Dynamic Folder)

Example

python train.py +sweep=optimization

Debug

python train.py +sweep=optimization +debug=sweep

Results

Example

python train.py +result=baseline

Debug

python train.py +result=baseline +debug=result

Hydra Config Printing

Base config for the run:

python train.py --cfg job

Hydra meta configs with overrides:

python train.py +ablation=learning_rate --cfg hydra -p hydra.sweeper

Usefull Scripts

Sync the export folder from the server to the local machien:

rsync -avz --progress thesis:/home/borth/neural-poisson/export /Users/robinborth/Desktop

Dataset

Thingi10k

In order to download the dataset, go to the Thingi10k website to see the different models

You can then download the datasets under the following link

https://drive.google.com/file/d/1RlDvNiFLDRztN0zWqQxmeraRG-XXFHUT/view

Please put the dataset into the Thingi10k folder in the data folder.

Sync Folders between Machines

rsync -avz data/Thingi10k/ baselines:/home/borth/data/Thingi10k