Skip to content

About

BlobBoards (BMVC 2026): fiducial markers made of a dense, multi-scale field of Gaussian blobs, with a feature-based pipeline for joint detection, identification and pose. Median translation error of 3.6-5.0 mm, the highest detection rate, and 69% of boards detected under 50% occlusion.

Resources

Stars

1 star

Watchers

0 watching

Forks

Repository files navigation

BlobBoards.jl

CI Docs Coverage

BlobBoards is a fiducial marker system: printable boards covered in a dense, multi-scale field of Gaussian blobs, and a pipeline that finds them in images, identifies each one and estimates its pose from hundreds of blob correspondences. This package authors the boards (PDF, TIFF or PNG, each with a JSON sidecar that reproduces it) and finds and tracks them on the CPU or a GPU. It is the implementation of BlobBoards: Robust Markers for Accurate Pose (BMVC 2026).

Installation

BlobBoards requires Julia 1.13 or later.

The example below finds a board on the CPU, and imports one name from VisualGeometryToolkitCore, the stack package that owns it:

pkg> add BlobBoards VisualGeometryToolkitCore

Install for your task

GPU finding, plotting, training and the Python interface are package extensions: they load once their trigger packages are installed and loaded. Each line names exactly the packages its task needs. Calling an extension's function without them raises an error naming the line to run.

Task Install Example
author boards (pattern, PDF, PNG) pkg> add BlobBoards author boards
find boards on the CPU pkg> add BlobBoards VisualGeometryToolkitCore below
find boards on the GPU pkg> add BlobBoards CUDA cuDNN VisualGeometryToolkitCore, then once using CUDA; CUDA.set_runtime_version!(v"12.9") and restart Julia GPU
track boards in a video pkg> add BlobBoards tracking
plot or visualize pkg> add BlobBoards CairoMakie plots
train the descriptor network pkg> add BlobBoards Lux Optimisers Enzyme Reactant NNlib TensorBoardLogger Zygote training
use from Python in a clone: pixi run -e py python (the package instantiates its own Julia environment) Python
work in a clone julia --project=<member> -e 'using Pkg; Pkg.instantiate()' clone

Example

Render a board as an image, then find it in that image with no camera (one homography per board). The default descriptor network is fetched once from the BlobBoards Zenodo record on first use; every file goes to a temporary directory.

# Install row: find boards on the CPU
using BlobBoards
using VisualGeometryToolkitCore: is_map_inlier

# A small board rendered at 100 pixels per inch: the image to search
dir = mktempdir()
blob_board("A5"; blob_sizes = (4.0mm, 0.2short_side), seed = 0xBEEF, dir, output = Png(100dpi))
board_json = only(filter(endswith(".json"), readdir(dir; join = true)))
scene = load_board(board_json)                 # the board's uid and bundle
_, image = BlobBoards.load(BlobBoards.BoardBundle, board_json)         # the rendered sheet, a grey image

network = DescriptorNetwork(descriptor_network_path(); execution_provider = :cpu)
gallery = BoardGallery(joinpath(dir, "board_features.jld2"), network)
boards  = Dict(insert!(gallery, [board_json], network))   # uid => the board's described blobs
bundles = Dict(scene.uid => scene.bundle)

config   = FindBoardsConfig(geometry = HomographyGeometry())
accepted = find_boards(boards, network, image; board_bundles = bundles, config)

uid, H, quality, scored = only(accepted)      # the board's uid, its board → image homography
n_inliers = count(is_map_inlier, scored.weights)
uid == scene.uid                               # true

Getting Started authors, prints and loads a first board.

Documentation

BlobBoards is part of the Visual Geometry Toolkit, Julia packages for robust visual geometry from Kiel University.

Citing

If you use BlobBoards in your research, please cite:

@inproceedings{pritts2026blobboards,
  author    = {Pritts, James and Sittart, Till and Sauer, Hendrik and
               Jan{\ss}en, Silja and Seegr{\"a}ber, Felix and
               Nakath, David and K{\"o}ser, Kevin},
  title     = {{BlobBoards}: Robust Markers for Accurate Pose},
  booktitle = {British Machine Vision Conference (BMVC)},
  year      = {2026},
  eprint    = {2608.28830},
  archivePrefix = {arXiv},
}

License

MIT

About

BlobBoards (BMVC 2026): fiducial markers made of a dense, multi-scale field of Gaussian blobs, with a feature-based pipeline for joint detection, identification and pose. Median translation error of 3.6-5.0 mm, the highest detection rate, and 69% of boards detected under 50% occlusion.

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages