OP-LIMap

#mobilerobotics project ##Example-Car Factory

Installation

Cloning The Repo

This repository has multiple nested Git submodules, so the cloning procedure is slightly more complex than usual. To clone the entirety of the repository:

git clone https://github.com/RaghavM11/LIMap-Extension.git
cd LIMap-Extension/
git submodule update --init --recursive

Installation via pipenv

There's multiple ways to install this correctly, but Dylan likes pipenv and it seems to work okay for him.

Tested on:

• Ubuntu 20.04
• CMake version 3.17

The process is:

Via Installation Script

While it is not perfect by any means, we have attempted to write a comprehensive installation script. This script:

• Verifies that your cmake version is compatible with LIMap
• Recursively updates the submodules in case you forgot to do so in Cloning the Repo
• Attempts to find colmap and PoseLib via CMake, though this is easily the most brittle of the steps.
• Checks apt dependencies are installed for LIMap
  • The script does not download these so as to preserve the system's state.
• Installs asdf for downloading Python executables for virtual environments
• Installs pipenv for Python virtual environment management (a personal favorite of Dylan's)
• Installs the pipenv environment Dylan used for the project
  • This:
    • installs Python 3.9 via asdf
    • installs all LIMap Python dependencies
    • installs LIMap
• Finally, this script overrides a file in one of the third-party directories.
  • This is likely due to LIMap not specifying a version requirement for Hierarchical-Localization, as they updated to using pathlib.Path instead of os.path among other syntactic changes.

Manual Install

1. Follow LIMAP installation (apt packages included), up to the point where you need to start installing Python dependencies (minus opencv).
   1. This includes installing colmap, building poselib, etc.
      1. Don't install torch or torchvision yet but do apt-install opencv libraries and such.
   2. Make sure your cmake is 3.17 or greater, I downloaded 3.17 and prepended the binary's parent path to my PATH variable.
      1. THIS IS CRUCIAL!
2. Install pipenv with pip install pipenv
3. In the LIMap-Extension directory, create pipenv environment with pipenv install --python 3.9 --site-packages
   1. If you do not have Python 3.9 and don't have certain packages necessary to fetch new Python versions, you'll get an error. Just follow the instructions in the error to install packages for fixing this.
4. Install pytorch and torchvision with

pipenv install torch==1.13.1 torchvision==0.14.1 --index https://download.pytorch.org/whl/cu116

1. Install limap dependencies. It is very important that this is done from the LIMap-Extension directory and not the limap sub-directory.

pipenv install -r ./limap/requirements.txt

1. Install limap

pipenv install -vv -e ./limap

1. Attempt to run a job. If the job fails due to not being able to find a third party library that LIMap uses, do the following:

pipenv install -vv -e ./limap/third-party/[THIRD PARTY PACKAGE THAT WASN'T FOUND]

Installation via Anaconda

1. Setup Conda env. (should have miniconda installed)

conda create -n myenv python=3.9 

2. Activate conda

conda activate myenv

3. Install dependecies in conda env following Via Installation Script (mentioned above) --Change CUDA version to 11.8 [if using 12]
4. install package

python -m pip install -Ive ./limap 

4. To double check if the package is successfully installed:

python -c "import limap"

Setup

1. For above example the files can be found here

##todo script Downloader script for sample dataset uploaded to drive (small, ~100 images)

For running examples

To run line mapping using Fitnmerge (line mapping with available depth maps) on Hypersim:

python limap/runners/hypersim/fitnmerge.py --output_dir limap/outputs/quickstart_fitnmerge

To run Visualization of the 3D line maps after the reconstruction:

python limap/visualize_3d_lines.py --input_dir outputs/quickstart_triangulation/finaltracks \
                             # add the camera frustums with "--imagecols outputs/quickstart_triangulation/imagecols.npy"

For Hybrid Loaclization of Points and Lines using NN detectors

python limap/runners/tests/localization.py --data limap/runners/tests/data/localization/localization_test_data_stairs_1.npy