kapture 1.0.17

kapture: file format for SfM

Overview

Kapture is a pivot file format, based on text and binary files, used to describe SfM (Structure From Motion) and more generally sensor-acquired data.

It can be used to store sensor parameters and raw sensor data:

• cameras

• images

• lidar and other sensor data

As well as computed data:

• 2d features

• 3d reconstruction

Finally, many popular datasets can directly be downloaded using the convenient downloader!

Specifications

The format specification is detailed in the kapture format specifications document.

Example File Structure

This is an example file structure of a dataset in the kapture format.

my_dataset                 # Dataset root path
├─ sensors/                # Sensor data root path
│  ├─ sensors.txt          # list of all sensors with their specifications (e.g. camera intrinsics)
│  ├─ rigs.txt             # geometric relationship between sensors (optional)
│  ├─ trajectories.txt     # extrinsics (timestamp, sensor, pose)
│  ├─ records_camera.txt   # all records of type 'camera' (timestamp, sensor and path to image)
│  ├─ records_SENSOR_TYPE.txt # all records of type SENSOR_TYPE (other sensors, eg: 'magnetic', 'pressure'...)
│  └─ records_data/            # image and lidar data path
│     ├─ map/cam_01/00001.jpg  # image path used in records_camera.txt (example)
│     ├─ map/cam_01/00002.jpg
│     ├─ map/lidar_01/0001.pcd # lidar data path used in records_lidar.txt
│     ├─ query/query001.jpg    # image path used in records_camera.txt
│     ├─ ...
├─ reconstruction/
│  ├─ keypoints/                    # 2D keypoints files
│  │  ├─ keypoints.txt              # type of keypoint
│  │  ├─ map/cam_01/00001.jpg.kpt   # keypoints for corresponding image (example)
│  │  ├─ query/query001.jpg.kpt     # keypoints for corresponding image (example)
│  │  ├─ ...
│  ├─ descriptors/                  # keypoint descriptors files
│  │  ├─ descriptors.txt            # type of descriptor
│  │  ├─ map/cam_01/00001.jpg.desc  # descriptors for corresponding image (example)
│  │  ├─ query/query001.jpg.desc    # descriptors for corresponding image (example)
│  │  ├─ ...
│  ├─ ...
│  ├─ points3d.txt                  # 3D points of the reconstruction
│  ├─ observations.txt              # 2D/3D points corespondences
│  ├─ matches/                      # matches files.
│  │  ├─ map/cam_01/00001.jpg.overlapping/cam_01/00002.jpg.matches # example
│  │  ├─ ...

Software

The kapture format is provided with a Python library, as well as several conversion tools.

Install

pip install kapture

or see installation for more detailed instructions.

Using docker

Build the docker image:

# build the docker image : if you have already cloned the repository
docker build . -t kapture/kapture
# OR build the docker image directly from github
docker build git://github.com/naver/kapture -t kapture/kapture
# run unit tests
docker run -it --rm kapture/kapture python3 -m unittest discover -s /opt/src/kapture/tests

If you want to process your own data, you can bind directories between the host and the container using --volume or --mount option (See the docker documentation). The following example mounts /path/to/dataset/ from the host to /dataset inside the docker.

docker run -it \
    --rm \ # Automatically remove the container when it exits \
    --volume /path/to/dataset/:/dataset:ro \ #read only
    kapture/kapture

kapture Python library

See the tutorial for some examples using the kapture Python library.

kapture tools

In this repository, you will find a set of conversion tools to or from kapture format. Import results to kapture format, and conversely, export converts kapture data to other formats. Depending of the format, some data might not be converted, either because the other format does not support it (—) or because its was not implemented (⨉). Here is a table summarizing the conversion capabilities:

conversion capabilities

Format	← →	cam	rig	img	trj	gps	kpt	dsc	gft	p3D	obs	mch
colmap	import	✓	✓	✓	✓	⨉	✓	✓	—	✓	✓	(✓)
export	✓	✓	✓	✓	⨉	✓	✓	—	✓	✓	(✓)
openmvg	import	✓	—	✓	✓	⨉	—	—	—	—	—	—
export	✓	—	✓	✓	⨉	—	—	—	—	—	—
OpenSfM	import	✓	⨉	✓	✓	✓	✓	✓	—	✓	⨉	✓
export	✓	⨉	✓	✓	⨉	✓	—	✓	—	⨉	✓
bundler	import	✓	—	✓	✓	—	✓	—	—	✓	✓	—
image_folder	import	—	—	✓	—	—	—	—	—	—	—	—
image_list	import	✓	—	✓	—	—	—	—	—	—	—	—
nvm	import	✓	—	✓	✓	—	✓	—	—	✓	✓	—
IDL_dataset_cvpr17	import	✓	—	✓	✓	—	—	—	—	—	—	—
RobotCar_Seasons	import	✓	✓	✓	✓	—	✓	?	—	✓	✓	?
ROSbag cameras+trajectory	import	(✓)	(✓)	✓	✓	⨉	—	—	—	—	—	—
SILDa	import	✓	✓	✓	✓	—	—	—	—	—	—	—
virtual_gallery	import	✓	✓	✓	✓	—	—	—	—	—	—	—

conversion capabilities

• ✓: supported, (✓) partially supported, ⨉: not implemented, —: not supported by format.

• cam: handle camera parameters, eg. intrisics

• rig: handle rig structure.

• img: handle the path to images.

• trj: handle trajectories, eg. poses.

• kpt: handle image keypoints locations.

• dsc: handle image keypoints descriptors.

• gft: handle global image feature descriptors.

• p3D: handle 3D point clouds.

• obs: handle observations, ie. 3D-points / 2D keypoints correspondences.

• mch: handle keypoints matches.

kapture support in other packages

Local Features

• R2D2 local features can be directly generated in kapture format. See here

• D2-Net features can also be extracted in kapture format. See instructions here.

Global Features

• AP-GeM global feature extractor in kapture format: here

Datasets

The kapture package provides conversion tools for several data formats and datasets used in the domain. But it also provides a tool to download datasets already converted to kapture. See the kapture tutorial for instructions to use the dataset downloader.

Here is a list of datasets you can directly download in kapture format with the downloader tool:

• Datasets from the Long Term Visual Localization site:

  • Aachen Day Night v1.1

  • Extended CMU-Seasons

  • RobotCar Seasons v2

  • InLoc (without images)

  • SILDa Weather and Time of Day

• Virtual Gallery dataset

kapture-localization

Checkout kapture-localization, our toolbox which contains implementations for various localization related algorithms.

• mapping and localization pipelines with custom features

• mapping and localization pipelines with SIFT and vocabulary tree matching (default colmap pipeline)

• image retrieval benchmark (global sfm, local sfm, pose approximation)

Tutorial

See the kapture tutorial for a short introduction to:

• conversion tools

• using kapture in your code

• dataset download

Contributing

There are many ways to contribute to the kapture project:

• provide feedback and suggestions of improvements

• submit bug reports in the project bug tracker

• provide a dataset in kapture format that we can add to the downloader tool

• implement a feature or bug-fix for an outstanding issue

• add support of kapture format in other software packages (e.g. SfM pipelines…​), thus adding support for more datasets

• provide scripts to create data in kapture format (e.g. local/global feature extraction)

• propose a new feature and implement it

If you wish to contribute, please refer to the CONTRIBUTING page.

License

Software license is detailed in the LICENSE file.

Contact Us

You can contact us through GitHub, or at kapture at naverlabs + com