Hugging Face's logo

Datasets:
sebothetramp
/
MBZUAI-Campus

Tasks:
Image-to-3D
Robotics
Other
Modalities:
Video
Languages:
English
Size:
< 1K
Tags:
3D reconstruction
computer vision
NeRF
Libraries:
Datasets
License:
Dataset card Data Studio Files Community
1
MBZUAI-Campus / README.md
Sebastian Cavada
new things
7de9793
preview code
|
raw
history blame contribute delete
3.11 kB
metadata
language:
  - en
pretty_name: MBZUAI Campus Reconstruction Dataset
tags:
  - 3D reconstruction
  - computer vision
  - NeRF
license: mit
task_categories:
  - image-to-3d
  - robotics
  - other
description: >
  This dataset provides the necessary files and scripts to reconstruct the
  MBZUAI campus using COLMAP, GLOMAP, and NERFstudio. It contains preprocessed
  video sequences and metadata required for hierarchical 3D reconstruction.


  The dataset includes:

  - Raw video sequences

  - Preprocessed frames

  - Calibration and metadata

  - Reconstruction scripts


  The hierarchical reconstruction starts with a base structure, followed by
  incremental updates with additional sequences.
installation: >
  This repository relies on COLMAP, GLOMAP, and NERFstudio. Ensure these
  dependencies are installed before running the scripts.
steps: |
  1. **Download the Files**
     - Clone the repository and extract the provided dataset folders.

  2. **Process Individual Video Sequences**
     - Run the following command to preprocess the video sequences:
       ```bash
       python prepare_data.py
       ```

  3. **Reconstruct the Campus Hierarchically**
     - Run the `core.sh` script to reconstruct the base structure:
       ```bash
       ./core.sh
       ```
     - For additional scenes, copy and modify `run_template.sh` to match the scene name and file, then execute it step-by-step or as a whole.
use_cases: |
  This dataset is essential for:
  - 3D environment reconstruction
  - Research on NeRF-based scene reconstruction
  - Structure-from-Motion experiments
  - Hierarchical multi-view scene understanding

This is the repository to download and prepare the MBZUAI campus reconstruction

Installation

TL,DR --> run install.sh (if it's there)

Conda environment

conda create --name nerfstudio -y python=3.8
conda activate nerfstudio
pip install --upgrade pip

CUDA Dependencies

pip install torch==2.1.2+cu118 torchvision==0.16.2+cu118 --extra-index-url https://download.pytorch.org/whl/cu118

conda install -c "nvidia/label/cuda-11.8.0" cuda-toolkit
pip install ninja git+https://github.com/NVlabs/tiny-cuda-nn/#subdirectory=bindings/torch

NERFstudio install

pip install nerfstudio

COLMAP install

conda install -y conda-forge::colmap

GLOMAP install

conda install -y conda-forge::glomap

ffmpeg for processing images

conda install conda-forge::ffmpeg

Reproduce results

1. Pre-Process individual videos sequences 

python prepare_data.py

2. Reconstruct the campus hierarchically

first run the run_core.sh script. This will reconstruct a base structure, upon which all the other reconstructions will be added to. This is the most important one.

For the other scenes, for now just copy the run_template.sh file and modify according to the current scane name and file. Usually I run line by line to make sure everything is correct, but theoretically you could just run the entire file.