ShapeEstimator

Installation

1. Make sure you have an updated nvidia grapich card driver
2. Install Docker Community
3. Install NVIDIA Container Toolkit

Download network and demo images

Network	Demo data
Google Drive	Google Drive

Unzip network in ./demo/data/models/,

./demo/data/models/

Afterwards it should be

./demo/data/models/ShapeNetwork

Unzip demo in

./demo/database/

Afterwards it should be

./demo/database/sfm

Usage

Build container by running:

./demo/docker_build.sh

Perform shape estimation by running:

./demo/docker_run_demo.sh "dataset"

As example:

./demo/docker_run_demo.sh fountain

The reconstructions are stored in

./demo/data/predictions/unsupervised/0/

For each image four files will be generated. "_ground_truth" which contains either a sparse or dense ground truth, "_initial" which contains the solution when $z$ is zero, "_pred" contains the 3D reconstruction and "_pred_filtered" where points with shallow angle have been filtered out.

Add Your Own Images

Put images in a folder

/path/name/images/

Run

./utility/docker_colmap.sh /path/name/

to to find a sparse sfm solution using colmap and convert it to the expected format. The solutions are saved in

./demo/database/sfm/processed/

Run

./demo/docker_run_demo.sh name

to reconstruct the 3D structure,

Optional

Install vscode with docker plugin for easy in docker container development (optional)