Ollin

Third-party notices

Ollin's own source is under the MIT license (see LICENSE). It also bundles the third-party components listed below, each under its own license. This file is the aggregate record; the full license text and the original copyright headers are kept alongside each component's source.

This is distinct from the projects Ollin is merely inspired by or studied as a reference: those contribute no code and no license obligations (see ATTRIBUTION.md). The components here are actual source redistributed inside this repository.

libtess2

  • Used for: triangulating concave polygons and polygons with holes, behind the vector Shape/Contour fill path (drawShape).
  • Location in this repo: External/CLibtess2/
  • Upstream: https://github.com/memononen/libtess2
  • Version: commit 8dbd6483e920311a58c9af10a10beb278efebc36 (2025-10-15)
  • License: SGI Free Software License B, Version 2.0 — full text at External/CLibtess2/LICENSE.txt
  • Also shipped as: part of Sources/Ollin/Resources/WebExpander.wasm, the stroke and fill expander compiled to WebAssembly that a web page a sketch exports carries when its strokes and fills travel as points (Scripts/build-web-expander.sh builds it). The page's script carries the copyright notice and the license reference beside the module.

Copyright (C) [dates of first publication] Silicon Graphics, Inc. All Rights Reserved.

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice including the dates of first publication and either this permission notice or a reference to http://oss.sgi.com/projects/FreeB/ shall be included in all copies or substantial portions of the Software.

(The software is provided "as is", without warranty of any kind. See LICENSE.txt for the full text, including the warranty disclaimer and the trademark clause.)

Box2D

  • Used for: the rigid-body solver behind OllinPhysics — bodies with rotation, polygon colliders, joints, and stable stacking — wrapped behind Ollin's own World/Body API.
  • Location in this repo: External/CBox2D/
  • Upstream: https://github.com/erincatto/box2d
  • Version: v3.1.1 — commit 8c661469c9507d3ad6fbd2fea3f1aa71669c2fe3
  • License: MIT — full text at External/CBox2D/LICENSE

MIT License

Copyright (c) 2022 Erin Catto

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction… (see LICENSE for the full text, including the warranty disclaimer).

Jolt Physics

  • Used for: the 3D rigid-body solver behind OllinPhysics (bodies with full 3D rotation, box/sphere/capsule/cylinder/hull/mesh colliders, joints, stacking, and grabbing), wrapped behind Ollin's own World3D/Body3D API.
  • Location in this repo: External/CJolt/
  • Upstream: https://github.com/jrouwe/JoltPhysics
  • Version: v5.6.0, commit e77f175595e64cb44218cc9d9d56fc365ad0e36a (2026-08-04)
  • License: MIT, full text at External/CJolt/LICENSE

Copyright 2021 Jorrit Rouwe

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction… (see LICENSE for the full text, including the warranty disclaimer).

Clipper2

  • Used for: polygon clipping and offsetting, behind Shape's boolean set operations (union, intersection, subtracting, symmetricDifference) and Shape.offset(by:join:) — wrapped behind Ollin's own API.
  • Location in this repo: External/CClipper2/
  • Upstream: https://github.com/AngusJohnson/Clipper2
  • Version: v2.0.1 — tag Clipper2_2.0.1, commit 21ebba05db8894f0c7217ad35ea518080f324946 (2026-06-10)
  • License: Boost Software License 1.0 — full text at External/CClipper2/LICENSE

Boost Software License - Version 1.0 - August 17th, 2003

Permission is hereby granted, free of charge, to any person or organization obtaining a copy of the software and accompanying documentation covered by this license (the "Software") to use, reproduce, display, distribute, execute, and transmit the Software… (see LICENSE for the full text, including the warranty disclaimer.)

Syphon Framework

  • Used for: sharing live GPU frames with other apps on the Mac (openFrameworks via ofxSyphon, Resolume, MadMapper, VDMX, …), behind OllinSyphon's SyphonServer/SyphonClient API. Only the Metal portion is vendored; the OpenGL path is omitted.
  • Location in this repo: External/CSyphon/
  • Upstream: https://github.com/Syphon/Syphon-Framework
  • Version: commit 71351d4b484cd2d1917867f7846a5cdca724552d (2025-10-06)
  • License: BSD 2-Clause — full text at External/CSyphon/License.txt
  • Local changes: the framework-style <Syphon/…> imports were rewritten to quoted includes for the flat SwiftPM target; SyphonServerRendererMetal.m was changed to compile its (unchanged) blit shader from embedded source at runtime instead of loading a precompiled metallib from a bundle (the swift run build produces no metallib); and SyphonMetalClient.m's frame texture gains MTLTextureUsagePixelFormatView so the consumer can read the surface through an sRGB view. and each .m gained one import of OllinSyphonPrefix.h, a file of ours that does what upstream's prefix header did, because forcing a prefix header in needs an unsafe build flag and a target carrying one cannot be reached through a package depended on by version. All are documented in External/CSyphon/README.md; the per-file copyright headers are intact.

Copyright the Syphon Project contributors — bangnoise (Tom Butterworth), vade (Anton Marini), Maxime Touroute & Philippe Chaurand. All rights reserved. See License.txt and the per-file headers for the exact notices.

Redistribution and use in source and binary forms, with or without modification, are permitted provided that the conditions of the BSD 2-Clause license are met (see License.txt for the full text, including the warranty disclaimer).

Hosek-Wilkie sky model

  • Used for: the procedural-sky environment (Environment.sky(...)). Its coefficient dataset and per-channel configuration code (RGB path) are cooked once on the CPU; Ollin's own Metal shader evaluates the sky radiance per texel from the result. Wrapped behind Ollin's own API.
  • Location in this repo: External/CHosekWilkie/, holding the upstream header and RGB dataset verbatim, with ArHosekSkyModel.c trimmed to the RGB path and the solar radiance function, and the spectral dataset trimmed to its solar and limb-darkening tables (ArHosekSkyModelData_Solar.h; the spectral/CIE sky models and their large datasets removed; see the directory's README.md).
  • Upstream: https://github.com/mmp/pbrt-v3 (src/ext), the reference implementation by Lukas Hosek and Alexander Wilkie (Charles University), v1.4a. Papers: "An Analytic Model for Full Spectral Sky-Dome Radiance" (SIGGRAPH 2012) and "Adding a Solar-Radiance Function to the Hosek-Wilkie Skylight Model" (IEEE CG&A 2013).
  • License: 3-clause BSD, full text at External/CHosekWilkie/LICENSE

Copyright (c) 2012 - 2013, Lukas Hosek and Alexander Wilkie. All rights reserved.

Redistribution and use in source and binary forms, with or without modification, are permitted provided that the copyright notice, the list of conditions, and the disclaimer are retained, and the contributors' names are not used to endorse derived products without permission. (The software is provided "as is"; see LICENSE for the full text and warranty disclaimer.)

MikkTSpace

  • Used for: generating vertex tangents (Mesh.tangents) for normal-mapped meshes that don't carry authored ones: the tangent-space standard the glTF 2.0 spec names for exactly that case, and the basis normal-map bakers target so a generated basis matches a baked texture. Wrapped behind Ollin's own API; the C symbols are not part of Ollin's public surface.
  • Location in this repo: External/CMikkTSpace/
  • Upstream: https://github.com/mmikk/MikkTSpace, by Morten S. Mikkelsen
  • Version: commit 3e895b49d05ea07e4c2133156cfa94369e19e409 (vendored 2026-08-11)
  • License: zlib-style per-file notice (upstream ships no standalone LICENSE file); reproduced at External/CMikkTSpace/LICENSE.txt and kept intact in both source files, as the notice requires.

Copyright (C) 2011 by Morten S. Mikkelsen

This software is provided 'as-is', without any express or implied warranty. Permission is granted to anyone to use this software for any purpose, including commercial applications, and to alter it and redistribute it freely, subject to the notice's three conditions (origin not misrepresented, altered versions marked, notice retained; see LICENSE.txt for the full text).

Spectral primaries data (simple-spectral)

  • Used for: spectral color (Spectrum, .paint mixing, and the spectral thin-film / diffraction / paint-mix GPU passes): the three BT.709 reflectance basis spectra a color decomposes into, plus the CIE 1931 2° observer and D65 illuminant tables, all at 380-780 nm in 5 nm steps. Cooked once on the CPU; Ollin's own shaders receive derived per-tap constants and never see the tables.
  • Location in this repo: External/CSpectralData/, the CSV data reformatted verbatim as C arrays (provenance and the exact files taken in the directory's README.md)
  • Upstream: https://github.com/geometrian/simple-spectral (commit 4b36e4d, 2020-07-19), the reference implementation accompanying Spectral Primary Decomposition for Rendering with sRGB Reflectance (Agatha Mallett and Cem Yuksel, EGSR 2019), by the paper's first author.
  • License: MIT, full text at External/CSpectralData/LICENSE

MIT License

Copyright (c) 2019

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction… (see External/CSpectralData/LICENSE for the full text, including the warranty disclaimer).

Cozette

MIT License

Copyright (c) 2020 Samhain &lt;samhain@moonwit.ch&gt; & contributors &lt;https://github.com/the-moonwitch/Cozette/contributors&gt;

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction… (see Cozette-LICENSE.txt for the full text, including the warranty disclaimer).

Hershey fonts

  • Used for: the bundled default stroke (single-line) font (StrokeFont.builtin, "Hershey Sans" / futural), loaded at runtime from its .jhf by the stroke-font parser and drawn by drawText.
  • Location in this repo: Sources/Ollin/Resources/futural.jhf (provenance in Sources/Ollin/Resources/Hershey-NOTICE.txt)
  • Upstream: the Hershey font data in James Hurt's .jhf form, as widely mirrored (e.g. https://github.com/kamalmostafa/hershey-fonts and https://paulbourke.net/dataformats/hershey/)
  • License: the glyph data is in the public domain. The .jhf distribution may be used by anyone for any purpose, provided two acknowledgments travel with the font data and the data is never converted to the format the U.S. NTIS distributes. Ollin ships the .jhf form unchanged in format.

The acknowledgments the distribution requires, as it words them:

The Hershey Fonts were originally created by Dr. A. V. Hershey while working at the U. S. National Bureau of Standards.

The format of the Font data in this distribution was originally created by James Hurt, Cognition, Inc., 900 Technology Park Drive, Billerica, MA 01821 (mit-eddie!ci-dandelion!hurt).

A note on the history: Hershey drew the faces at the U.S. Naval Weapons Laboratory in Dahlgren, Virginia, around 1967 (NWL Report 2101, Calligraphy for Computers), and the National Bureau of Standards published the data in 1976. The first acknowledgment is kept as the distribution requires it.

Blender matcaps

  • Used for: the bundled built-in matcap (material-capture) sphere textures (Matcap.chrome, .clay, .toon, …), sampled by the view-space normal by matcap(_:) to shade a 3D mesh.
  • Location in this repo: Sources/Ollin/Resources/Matcaps/ (per-file mapping + provenance in Sources/Ollin/Resources/Matcaps/Matcaps-LICENSE.txt)
  • Upstream: the matcaps bundled with Blender — https://projects.blender.org/blender/blender/src/branch/main/release/datafiles/studiolights/matcap (each source .exr was combined from its diffuse + specular passes into a single sRGB PNG at native 512×512)
  • License: CC0 1.0 / public domain — per the upstream license.txt

These matcap images are licensed as CC0 or public domain. Thanks to the Blender community for contributing these matcaps. CC0 carries no attribution requirement; the credit above is given freely, and provenance is recorded for the bundled assets.

Bundled HDRI environments

  • Used for: the built-in HDRI environment maps that light 3D surfaces through image-based lighting. Eight are bundled in this repo (Environment.studio, .city, .courtyard, .forest, .interior, .night, .sunrise, .sunset), loaded at runtime. A curated dozen more (.day, .dusk, .snow, …), plus any highRes(_:) 2K/4K/8K upgrade and any user hdri(downloadURL:), download on demand from Poly Haven and are cached on the user's machine, not redistributed in this repo.
  • Location in this repo: Sources/Ollin/Resources/Environments/ (the eight bundled files; per-file mapping + provenance in Sources/Ollin/Resources/Environments/Environments-LICENSE.txt)
  • Upstream: two CC0 sources. Blender's bundled "world" studiolights (https://projects.blender.org/blender/blender/src/branch/main/release/datafiles/studiolights/world), themselves created by Greg Zaal from Poly Haven; and Poly Haven directly (https://polyhaven.com/hdris), by Greg Zaal and contributors. Each bundled file was resampled to 1024×512 and saved as a half-float PIZ OpenEXR (the most compact HDR format ImageIO decodes natively); the on-demand downloads come from Poly Haven at their native resolution.
  • License: CC0 1.0 / public domain, per the upstream license pages (Blender world, Poly Haven).

These HDRIs are licensed as CC0 / public domain. Thanks to Greg Zaal, Poly Haven, and the Blender community. CC0 carries no attribution requirement; the credit above is given freely, and provenance is recorded for the bundled assets.

LTC lookup tables

  • Used for: area-light shading (rectLight / diskLight / tubeLight): the fitted linearly-transformed-cosine tables the lit-mesh fragments sample to approximate the GGX response of a glowing rectangle, disk, or tube.
  • Location in this repo: Sources/Ollin/Resources/LTC/ltc_tables.bin (conversion recipe + provenance in Sources/Ollin/Resources/LTC/LTC-NOTICE.txt)
  • Upstream: https://github.com/selfshadow/ltc_code (commit 31e5e96), the reference implementation accompanying Real-Time Polygonal-Light Shading with Linearly Transformed Cosines (Eric Heitz, Jonathan Dupuy, Stephen Hill, and David Neubelt, SIGGRAPH 2016). The two 64×64 float tables from fit/results/ltc.js, repacked as raw little-endian float32 with no reordering.
  • License: BSD 3-Clause; full text at Sources/Ollin/Resources/LTC/LTC-NOTICE.txt

Copyright (c) 2017, Eric Heitz, Jonathan Dupuy, Stephen Hill and David Neubelt. All rights reserved.

Redistribution and use in source and binary forms, with or without modification, are permitted provided that the conditions in the upstream license are met, including reproducing the copyright notice and referencing the paper… (see LTC-NOTICE.txt for the full text, including the warranty disclaimer).

Marble Madness (example font)

  • Used for: the PlaydateFont example sketch only — a sample Playdate .fnt font, loaded at runtime to demonstrate the loader. Not part of the Ollin framework; Ollin bundles no .fnt fonts itself.
  • Location in this repo: Examples/Text/PlaydateFont/MarbleMadness.fnt
  • Upstream: https://github.com/idleberg/playdate-arcade-fonts (an original homage to classic arcade typography)
  • License: CC0 1.0 (Public Domain Dedication) — https://creativecommons.org/publicdomain/zero/1.0/

The fonts in playdate-arcade-fonts are released into the public domain under CC0, which carries no attribution requirement; the credit above is given freely.

El Fandanguito (example audio)

  • Used for: the FilePlayer and SoundReactive example sketches only: a sample audio clip, to demonstrate AudioPlayer reacting to a file and Soundtrack analyzing a video's audio. Not part of the Ollin framework; Ollin (and OllinAudio) bundle no audio themselves.
  • Location in this repo: Examples/Audio/FilePlayer/fandanguito.m4a, with a second copy beside the Tuner example at Examples/Audio/Tuner/fandanguito.m4a; it is also the audio track of Examples/Video/SoundReactive/voladores-fandanguito.mp4 (see the Voladores de Papantla entry)
  • Work: El Fandanguito, a traditional Mexican son huasteco (the composition is traditional / public domain). Performed on violin by Cynthia Molina; recorded and edited by Wikimedia Commons users Emropa and ClawisJM (students of Tec de Monterrey).
  • Upstream: https://commons.wikimedia.org/wiki/File:Viol%C3%ADn_SonHuasteco_ELFandanguito.ogg
  • License: CC BY-SA 4.0 — https://creativecommons.org/licenses/by-sa/4.0

Licensed CC BY-SA 4.0 (Attribution-ShareAlike). Attribution is given above as required. Changes: the original Ogg Vorbis file was transcoded to AAC (with short fades) for AVFoundation playback; no other edits. As a ShareAlike work this clip remains under CC BY-SA 4.0, and that obligation applies to the audio file and its adaptations, not on Ollin's source, which stays MIT (the clip is merely bundled alongside it). The root LICENSE is unaffected.

Voladores de Papantla (example video)

  • Used for: the VideoPlayback, FootageDepth and SoundReactive example sketches only: a sample video clip, to demonstrate VideoPlayer drawing a file as a live image, DepthClip reading a recording's depth whole, and a soundtrack analyzer reacting to it. Not part of the Ollin framework; Ollin (and OllinVideo) bundle no video themselves.
  • Location in this repo: Examples/Video/VideoPlayback/voladores.mp4, with an identical copy at Examples/Vision/FootageDepth/voladores.mp4; Examples/Video/SoundReactive/voladores-fandanguito.mp4 is a further-trimmed excerpt paired with the El Fandanguito recording below as its soundtrack (the source video's own audio track is silent)
  • Work: Voladores de Papantla México — a recording of the Danza de los Voladores, the Totonac pole-flying ritual dance from Papantla, Veracruz (performed at an exhibition in Mexico City, 2018). Filmed by José Millán (Wikimedia Commons user Jmillan325).
  • Upstream: https://commons.wikimedia.org/wiki/File:Voladores_de_Papantla_M%C3%A9xico.webm
  • License: CC BY-SA 4.0 — https://creativecommons.org/licenses/by-sa/4.0

Licensed CC BY-SA 4.0 (Attribution-ShareAlike). Attribution is given above as required. Changes: a ~28-second excerpt was trimmed from the original 3-minute VP9/WebM file and re-encoded to H.264 MP4 at 960×540 (mono AAC audio) for AVFoundation playback and a small repository footprint; no other edits. The SoundReactive variant is a further 16-second excerpt of that trim, muxed with the El Fandanguito recording (see its own entry) as the audio track. As a ShareAlike work this clip remains under CC BY-SA 4.0: that obligation applies to the video file and its adaptations, not on Ollin's source, which stays MIT (the clip is merely bundled alongside it). The root LICENSE is unaffected.

Sample photographs (example assets)

  • Used for: the OllinSamplePhotos library (photographs and one short film), which the Images, Color, and Effects example sketches and the Guide's and reference's figures read so a filter, a mosaic, a dither, or a winding can be shown on a real picture. A product of its own, so an app that never imports it ships none of them; the Ollin framework itself bundles no photograph.
  • Location in this repo: Sources/OllinSamplePhotos/Resources/, twenty JPEGs and one film, re-encoded at quality 72 to 95 with the metadata stripped. The eighteen photographs are 1600 pixels on the long side: four faces cropped square from their originals, four whole figures kept in whatever frame holds the body, two tables seen from above, cropped square, four streets cropped to the largest square each original held, two landscapes at dusk kept wide at three to two, and a book page and a pair of open hands in their whole frames. The two surface textures are 1024-pixel squares, the size a GPU mips evenly. Each credit is also carried in code as SamplePhoto.credit.
  • portrait.jpg: Vibrant portrait of a woman in traditional Oaxaca attire by Jhovani Morales, Oaxaca de Juárez, Mexico. https://www.pexels.com/photo/13402074/ (Pexels License).
  • scarf.jpg: An elderly woman with a scarf looks directly at viewer by Matthew Stephenson, Oaxaca, Mexico. https://unsplash.com/photos/j6OYgSMVuiI — Unsplash License.
  • profile.jpg: Profile of a woman with a braided ponytail against a solid tan background by Jessica Felicio, Berrien County, United States. https://unsplash.com/photos/QS9ZX5UnS14 — Unsplash License.
  • marigolds.jpg: Portrait of an elderly Mexican woman with vibrant cempasuchil flowers for Día de Muertos celebration by Fernando Paleta, Mexico City, Mexico. https://www.pexels.com/photo/29302608/ (Pexels License).
  • reaching.jpg: Man Dancing Alone by Los Muertos Crew, Mexico. https://www.pexels.com/photo/8854016/ (Pexels License).
  • wrestler.jpg: Masked wrestler triumphs in lively arena by Juan TM, Santiago de Querétaro, Mexico. https://www.pexels.com/photo/30098566/ (Pexels License).
  • dancer.jpg: Graceful contemporary dancer posing barefoot in a flowing black skirt by Gustavo Fring. https://www.pexels.com/photo/7447230/ (Pexels License).
  • handstand.jpg: Energetic breakdancer in urban setting performing a handstand mid-move by Gabriel Jiménez. https://www.pexels.com/photo/26870547/ (Pexels License).
  • breakfast.jpg: Top view of a vibrant Oaxacan breakfast with various dishes on a wooden table by Jorge Acre, Oaxaca de Juárez, Mexico. https://www.pexels.com/photo/17061842/ (Pexels License).
  • desk.jpg: flat-lay photography of MacBook, coffee-filled cup, book, and wristwatch by Ben Kolde, Oxford, United States. https://unsplash.com/photos/H29h6a8j8QM (Unsplash License).
  • alley.jpg: Colorful Guanajuato Street Under Blue Sky by Emanuel Estrada, Guanajuato, Mexico. https://www.pexels.com/photo/33880220/ (Pexels License).
  • street.jpg: Colorful Decorations over Street by Jorge Acre, San Miguel de Allende, Mexico. https://www.pexels.com/photo/16625366/ (Pexels License).
  • textiles.jpg: multicolored textiles lot by analuisa gamboa, Teotitlán del Valle, Oaxaca, Mexico. https://unsplash.com/photos/0ohjyDUIUq0 (Unsplash License).
  • city.jpg: people walking on street near brown concrete building during daytime by Gerardo Martin Fernandez Vallejo, Guanajuato, Mexico. https://unsplash.com/photos/ukIew--AEOc (Unsplash License).
  • dance.mp4: Video of a Man Dancing by Antoni Shkraba. https://www.pexels.com/video/video-of-a-man-dancing-7571381/ (Pexels License). Cut to its first thirty seconds and to a square, scaled to 960, sound removed.
  • hands.jpg: Photography of Left and Right Palms by Luis Quintero. https://www.pexels.com/photo/2258247/ (Pexels License).
  • page.jpg: Shadows on Open Book by Saliha Öner. https://www.pexels.com/photo/8466090/ (Pexels License).
  • talavera.jpg: Traditional tin glazed tilework with colorful pattern by rotekirsche 20. https://www.pexels.com/photo/5438689/ (Pexels License).
  • stone.jpg: Textured Stone Wall Surface Close-Up by Memet Öz. https://www.pexels.com/photo/36023243/ (Pexels License).
  • headland.jpg: Silhouette of Mountain Near Body of Waters by Joe Leineweber, Loreto, Baja California Sur, Mexico. https://www.pexels.com/photo/12848727/ (Pexels License).
  • boats.jpg: boats in the water by Elesban Landero Berriozábal, Cozumel, Quintana Roo, Mexico. https://unsplash.com/photos/ACQmpRYafPg (Unsplash License).
  • Licenses: Unsplash License (https://unsplash.com/license); Pexels License (https://www.pexels.com/license/)

Both licenses grant a free, worldwide license to download, copy, modify, and use the photographs, commercial use included, with no attribution required; the photographers are credited here and in code all the same. Both forbid selling unaltered copies and compiling the pictures into a competing stock service, and the Pexels License adds that an identifiable person may not be shown in a bad light or as endorsing anything, which is how they are used: as the picture under a filter, a screen, a winding, or a pose skeleton. Changes: each was resized to 1600 pixels on its long side, converted to sRGB, and re-encoded as JPEG with its metadata (location included) removed. The four faces were also cropped square around the subject, and reaching was cropped to bring the figure up in the frame; the other three figures keep the whole frame the photographer shot. Both tables were cropped square, the desk cropped in far enough that the print on the notebook can be read. No other edits. The desk photograph carries a notebook whose printed cover names its maker, which the text recognizer reads aloud in the TextScan example; that is incidental to a photograph of a desk and is not an endorsement. The licenses apply to the photographs, not on Ollin's code, which stays MIT. The root LICENSE is unaffected.

Video Depth Anything, small (example model, built locally, not bundled)

  • Used for: the DepthContours and FootageDepth example sketches and the DepthTracker and DepthClip they demonstrate: temporally consistent video depth over the live feed, and over a whole recording read ahead of time. Not part of the Ollin framework; Ollin (and OllinVision) bundle no model weights themselves.
  • Location in this repo: none. Nobody publishes a Core ML build of this model, so Scripts/fetch-models.sh makes one on a developer's machine: Scripts/convert-video-depth.sh clones the upstream repository at a pinned commit, downloads the checkpoint, and Scripts/convert-video-depth.py traces the model's streaming step into VideoDepthAnythingSmallF16.mlpackage and its 32-frame window into VideoDepthAnythingSmallClipF16.mlpackage (each checked against the upstream code), and writes VideoDepthClipReference.bin, the upstream clip inference over a fixed synthetic clip that the tests check the Swift scheduler against, all under the gitignored Models/. The upstream source is used only at conversion time, in that work directory; none of it is copied into this repository.
  • Work: Video Depth Anything: Consistent Depth Estimation for Super-Long Videos (the small checkpoint, video_depth_anything_vits.pth), Sili Chen, Hengkai Guo, Shengnan Zhu, Feihu Zhang, Zilong Huang, Jiashi Feng, Bingyi Kang (CVPR 2025), https://arxiv.org/abs/2501.12375. Its temporal head builds on the AnimateDiff motion module (Yuwei Guo et al., Apache-2.0), which the upstream code credits in its own headers.
  • Upstream: https://github.com/DepthAnything/Video-Depth-Anything (code, Apache-2.0) and https://huggingface.co/depth-anything/Video-Depth-Anything-Small (the checkpoint).
  • License: Apache-2.0, https://www.apache.org/licenses/LICENSE-2.0 (the code and the small checkpoint; the Base and Large checkpoints are CC BY-NC 4.0 and are not used)

Apache-2.0 applies to the checkpoint the script downloads and to the package it derives from it; nothing from the model is redistributed in this repository, so the root LICENSE is unaffected. The conversion script is Ollin's own work (MIT with the rest of the repository), written against the model's published inference, streaming and windowed.

Depth Anything V2, small (example model — downloaded, not bundled)

  • Used for: the DepthRelief example sketch only — a monocular depth-estimation model, to demonstrate ModelTracker running a custom Core ML model. Not part of the Ollin framework; Ollin (and OllinVision) bundle no model weights themselves.
  • Location in this repo: none. The weights are not committed: Scripts/fetch-models.sh downloads them into the gitignored Models/ directory on a developer's machine.
  • Work: Depth Anything V2 (the small checkpoint), Lihe Yang, Bingyi Kang, Zilong Huang, Zhen Zhao, Xiaogang Xu, Jiashi Feng, Hengshuang Zhao (2024) — in Apple's official Core ML conversion (DepthAnythingV2SmallF16.mlpackage).
  • Upstream: https://huggingface.co/apple/coreml-depth-anything-v2-small (conversion), via Apple's model gallery https://developer.apple.com/machine-learning/models/ — original model https://github.com/DepthAnything/Depth-Anything-V2
  • License: Apache-2.0 — https://www.apache.org/licenses/LICENSE-2.0 (the small checkpoint; the larger Depth Anything V2 checkpoints are CC BY-NC 4.0 and are not used)

Apache-2.0 applies to the model weights the script downloads; nothing from the model is redistributed in this repository, so the root LICENSE is unaffected. This entry records the provenance of what the script fetches.

YOLOv3-tiny (example model — downloaded, not bundled)

  • Used for: the ObjectDetection example sketch only — an object-detection model (80 COCO classes), to demonstrate ModelTracker's labeled-box surface. Not part of the Ollin framework; Ollin (and OllinVision) bundle no model weights themselves.
  • Location in this repo: none. The weights are not committed: Scripts/fetch-models.sh downloads them into the gitignored Models/ directory on a developer's machine.
  • Work: YOLOv3-tiny, Joseph Redmon and Ali Farhadi — "YOLOv3: An Incremental Improvement" (2018) — in Apple's Core ML conversion (YOLOv3TinyFP16.mlmodel).
  • Upstream: https://developer.apple.com/machine-learning/models/ (conversion) — original model https://github.com/pjreddie/darknet, https://pjreddie.com/darknet/yolo/
  • License: YOLO License, Version 2 (a public-domain dedication: "Darknet is public domain. Do whatever you want with it.") — https://github.com/pjreddie/darknet/blob/master/LICENSE

The YOLO License applies to the model the script downloads; nothing from the model is redistributed in this repository, so the root LICENSE is unaffected. This entry records the provenance of what the script fetches.

MNIST drawing classifier (example model — downloaded, not bundled)

  • Used for: the DigitReader example sketch only — a handwritten-digit classifier, to demonstrate ModelTracker reading a sketch's own pixel-authored drawing through the still detect(in:) path. Not part of the Ollin framework; Ollin (and OllinVision) bundle no model weights themselves.
  • Location in this repo: none. The weights are not committed: Scripts/fetch-models.sh downloads them into the gitignored Models/ directory on a developer's machine.
  • Work: MNISTClassifier — Apple's Turi Create-trained drawing classifier from the Core ML model gallery, trained on the MNIST dataset of handwritten digits (LeCun, Cortes, Burges).
  • Upstream: https://developer.apple.com/machine-learning/models/ — dataset http://yann.lecun.com/exdb/mnist/
  • License: MIT (Copyright 2019 Apple Inc.) — https://docs-assets.developer.apple.com/coreml/models/Image/DrawingClassification/MNISTClassifier/LICENSE-MIT.txt

MIT applies to the model the script downloads; nothing from the model is redistributed in this repository, so the root LICENSE is unaffected. This entry records the provenance of what the script fetches. (The StyleMirror example needs no entry here at all: its style-transfer model is trained by the user — Scripts/train-style-model.swift, over the CreateML framework — so the weights are the user's own work; nothing is fetched or redistributed.)

DeepLabV3 (example model — downloaded, not bundled)

  • Used for: the PaintByClass example sketch only — a semantic-segmentation model (21 PASCAL VOC classes), to demonstrate ModelTracker's class-mask surface. Not part of the Ollin framework; Ollin (and OllinVision) bundle no model weights themselves.
  • Location in this repo: none. The weights are not committed: Scripts/fetch-models.sh downloads them into the gitignored Models/ directory on a developer's machine.
  • Work: DeepLabV3 (MobileNetV2 backbone), Liang-Chieh Chen, Yukun Zhu, George Papandreou, Florian Schroff, Hartwig Adam — "Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation" (2018) — in Apple's Core ML conversion (DeepLabV3FP16.mlmodel).
  • Upstream: https://developer.apple.com/machine-learning/models/ (conversion) — original model https://github.com/tensorflow/models/tree/master/research/deeplab (TensorFlow)
  • License: Apache-2.0 — https://github.com/tensorflow/models/blob/master/LICENSE (per the model's own embedded license metadata, which points at the TensorFlow repositories)

Apache-2.0 applies to the model weights the script downloads; nothing from the model is redistributed in this repository, so the root LICENSE is unaffected. This entry records the provenance of what the script fetches.

MobileCLIP-S0 (example model, downloaded, not bundled)

  • Used for: ConceptTracker and its TugOfWords example sketch. The framework provides no bundled weights.
  • Location in this repo: none. Scripts/fetch-models.sh downloads the weights into the gitignored Models/ directory.
  • Work: MobileCLIP-S0 (CVPR 2024), by Pavan Kumar Anasosalu Vasu, Hadi Pouransari, Fartash Faghri, Raviteja Vemulapalli, and Oncel Tuzel, in Apple's Core ML exports (mobileclip_s0_image.mlpackage, mobileclip_s0_text.mlpackage).
  • Upstream: https://huggingface.co/apple/coreml-mobileclip (export); original release https://github.com/apple/ml-mobileclip
  • License: the 2024 release's LICENSE_weights_data: redistributable with attribution; no research-only clause. The export's model card declares apple-ascl. The fetch script uses the 2024 model (MobileCLIP2 is research-only).

The license applies to the downloaded weights. This repository does not redistribute them; the root LICENSE is unaffected.

Segment Anything 2.1 (example model, downloaded, not bundled)

  • Used for: PointSegmenter and its PointLift example sketch. The framework provides no bundled weights.
  • Location in this repo: none. Scripts/fetch-models.sh downloads the weights into the gitignored Models/ directory.
  • Work: SAM 2: Segment Anything in Images and Videos (2024), Nikhila Ravi, Valentin Gabeur, Yuan-Ting Hu, Ronghang Hu, Chaitanya Ryali, Tengyu Ma, Haitham Khedr, Roman Rädle, Chloe Rolland, Laura Gustafson, Eric Mintun, Junting Pan, Kalyan Vasudev Alwala, Nicolas Carion, Chao-Yuan Wu, Ross Girshick, Piotr Dollár, Christoph Feichtenhofer (Meta AI): the 2.1 "small" checkpoint, in Apple's official Core ML conversion, split as SAM2_1SmallImageEncoderFLOAT16.mlpackage, SAM2_1SmallPromptEncoderFLOAT16.mlpackage, and SAM2_1SmallMaskDecoderFLOAT16.mlpackage.
  • Upstream: https://huggingface.co/apple/coreml-sam2.1-small (conversion); original release https://github.com/facebookresearch/sam2
  • License: Apache-2.0 (both the original release and the Core ML conversion's model card).

Apache-2.0 applies to the model weights the script downloads; nothing from the model is redistributed in this repository, so the root LICENSE is unaffected. This entry records the provenance of what the script fetches.

CLIP byte-pair-encoding vocabulary (example data, downloaded, not bundled)

  • Used for: ConceptTracker's tokenizer (PhraseTokenizer), which turns a typed phrase into the token sequence the text encoder expects. The tokenizer implementation is Ollin's own, credited in ATTRIBUTION.md; this entry covers the vocabulary file.
  • Location in this repo: none. Scripts/fetch-models.sh downloads bpe_simple_vocab_16e6.txt.gz into the gitignored Models/ directory and unpacks it.
  • Work: the byte-pair-encoding merges file from OpenAI's CLIP tokenizer. MobileCLIP's text encoder was trained on this vocabulary.
  • Upstream: https://github.com/openai/CLIP (clip/bpe_simple_vocab_16e6.txt.gz)
  • License: MIT (Copyright 2021 OpenAI): https://github.com/openai/CLIP/blob/main/LICENSE

The license applies to the downloaded file. This repository does not redistribute it; the root LICENSE is unaffected.

↑↓ move ↩ open