NAPA

Not Another Pixel Algorithm: NAPA skips most of the pixels in each frame.

NAPA is a fast AprilTag detector. Give it a greyscale frame and it returns tag IDs and sub-pixel corners, plus poses with covariances if you give it a camera calibration. It reads every AprilTag family, with tag36h11, the FRC family, as the default.

python -m pip install napa-vision

Compared with AprilTag

These results compare NAPA with upstream AprilTag 3 at decimate 2 (WPILib's default) and decimate 1 (full resolution, PhotonVision's default). The benchmark is 3,000 rendered frames holding 5,521 tags, each with exact ground truth, and every detector ran single-threaded on the same CPU core.

NAPA was faster than both on every frame:

Range scales with the inverse of the smallest tag found, so on the same camera NAPA sees about 3.6× as far as decimate 2 and 1.6× as far as decimate 1.

NAPA's one wrong ID was a tag-shaped decoy whose random pattern fell within 2 bits of a real code. On tag16h5, the 2023 FRC family, AprilTag at its default two-bit correction reported 581 wrong IDs in the family benchmark, and NAPA reported none.

The frames are synthetic. To check NAPA on your own camera, score it on your own footage. The benchmark guide has the results for every scene, resolution, and tag family.

How it works

NAPA finds likely tags on a sparse grid of pixels and only does full-resolution work on the tags themselves. How NAPA works explains each step.

Quick start

Requires Python 3.11 or newer.

from napa_vision import NapaDetector

detector = NapaDetector()
for tag in detector.detect(image):  # uint8 greyscale, BGR, or BGRA
    print(tag.tag_id, tag.corners)

Code written for pupil_apriltags or dt-apriltags works after changing one import:

from napa_vision.compat import Detector  # was: from pupil_apriltags import Detector

For poses, install OpenCV with the headless extra and give NAPA a calibration:

python -m pip install "napa-vision[headless]"

from napa_vision import VisionProcessor, load_calibration_data

processor = VisionProcessor(calibration=load_calibration_data("camera.json"), tag_size_m=0.1651)
result = processor.process(frame, timestamp=capture_time)

Documentation

• Detection: tuning, tag families, printing tags, and switching from pupil_apriltags
• Poses and filters: metric poses, coordinate frames, and WPILib pose estimators
• Camera calibration
• Stereo rigs: one pose from two calibrated cameras
• Platforms: C and C++, Android and Meta Quest, and building from source
• Measuring on your own footage
• How NAPA works: the pipeline, obstructions, and limits
• API reference: every class, method, and field

Command-line tools

License

Copyright © 2026 Zachary Wilson. NAPA is free for noncommercial use under the PolyForm Noncommercial License 1.0.0: personal projects, research, education, student robotics teams, and other noncommercial organizations can use, modify, and share it. Copies, forks, and works based on NAPA must include the license and its notices crediting Zachary Wilson. Commercial use, including selling NAPA or products built on it, needs a separate license. Ask through the repository.

The tag families' codewords come from AprilTag under its BSD license. See THIRD_PARTY_NOTICES.