Initial ONNX drop: pinned mirrors of YuNet, ArcFace, CLIP ViT-B/32 image+text, CLIP BPE vocab

LICENSES.md

@@ -0,0 +1,40 @@
+# Licenses
+
+This repository mirrors third-party models. Each file remains under its
+upstream license, reproduced/linked below. Arca claims no ownership of any
+model here.
+
+## yunet.onnx — MIT
+
+Source: <https://github.com/opencv/opencv_zoo/tree/main/models/face_detection_yunet>
+(`face_detection_yunet_2023mar.onnx`). YuNet is by Shiqi Yu and Wei Wu.
+OpenCV Zoo states: "All files in this directory are licensed under MIT License."
+
+## arcface.onnx — Apache-2.0
+
+Source: <https://github.com/onnx/models/tree/main/validated/vision/body_analysis/arcface>
+(`arcfaceresnet100-8.onnx`), from the ONNX Model Zoo. ArcFace is by Deng et al.
+Licensed Apache-2.0 per the model's directory in the ONNX Model Zoo.
+
+## clip-vit-b32-image.onnx / clip-vit-b32-text.onnx — MIT
+
+Sources: <https://huggingface.co/Qdrant/clip-ViT-B-32-vision> and
+<https://huggingface.co/Qdrant/clip-ViT-B-32-text>. ONNX exports of OpenAI's
+CLIP ViT-B/32. Both upstream repos declare `license: mit`. Original CLIP is by
+OpenAI, released under MIT.
+
+## bpe_simple_vocab_16e6.txt.gz — MIT
+
+Source: <https://github.com/openai/CLIP> (`clip/bpe_simple_vocab_16e6.txt.gz`).
+OpenAI's CLIP repository is MIT-licensed.
+
+---
+
+## Deliberately NOT hosted here
+
+**YOLOv8n** (and the wider Ultralytics YOLO family) is **AGPL-3.0**. The
+`unity/inference-engine-yolo` mirror declares GPL-3.0. AGPL-3.0 requires that a
+larger work embedding these weights be released in full under AGPL-3.0, or that
+the user hold an Ultralytics Enterprise License. Arca is closed-source, so these
+weights are neither shipped nor mirrored. See
+<https://www.ultralytics.com/license>.

README.md

@@ -0,0 +1,64 @@
+---
+license: other
+license_name: mixed-permissive
+license_link: https://huggingface.co/umbml/arca-models/blob/main/LICENSES.md
+library_name: onnx
+tags:
+  - onnx
+  - clip
+  - face-detection
+  - face-recognition
+  - mirror
+---
+
+# arca-models
+
+Pinned ONNX mirrors of third-party models used for **on-device** inference in
+[Arca](https://arca.umbml.com). Nothing here is trained by us — every file is a
+byte-for-byte copy of an upstream release, re-hosted so that:
+
+1. **Downloads are immutable.** Arca pins each file by commit SHA and verifies a
+   SHA-256 before use. Fetching upstream `main` gave us a moving target: two of
+   our source URLs 404'd after upstream repos were restructured, silently
+   breaking on-device features.
+2. **Upstream repos aren't used as a CDN.** Pulling a 261 MB file from a
+   `github.com/.../raw/main/` URL on every client install is not what that
+   endpoint is for.
+
+If you want these models, please prefer the original sources below — they are
+the authoritative, maintained copies.
+
+## Contents
+
+| File | Bytes | Upstream source | License |
+| --- | --- | --- | --- |
+| `yunet.onnx` | 232,589 | [opencv/opencv_zoo](https://github.com/opencv/opencv_zoo/tree/main/models/face_detection_yunet) — `face_detection_yunet_2023mar.onnx` | MIT |
+| `arcface.onnx` | 261,036,388 | [onnx/models](https://github.com/onnx/models/tree/main/validated/vision/body_analysis/arcface) — `arcfaceresnet100-8.onnx` | Apache-2.0 |
+| `clip-vit-b32-image.onnx` | 351,686,194 | [Qdrant/clip-ViT-B-32-vision](https://huggingface.co/Qdrant/clip-ViT-B-32-vision) | MIT |
+| `clip-vit-b32-text.onnx` | 254,102,519 | [Qdrant/clip-ViT-B-32-text](https://huggingface.co/Qdrant/clip-ViT-B-32-text) | MIT |
+| `bpe_simple_vocab_16e6.txt.gz` | 1,356,917 | [openai/CLIP](https://github.com/openai/CLIP) | MIT |
+
+### SHA-256
+
+```
+8f2383e4dd3cfbb4553ea8718107fc0423210dc964f9f4280604804ed2552fa4  yunet.onnx
+f3a6bc281e72f88862f5748b53be3d76b3b48f8f1ab1f4a537941bdc4e1b01da  arcface.onnx
+c68d3d9a200ddd2a8c8a5510b576d4c94d1ae383bf8b36dd8c084f94e1fb4d63  clip-vit-b32-image.onnx
+4dbe762b11e36488304471e439cde89da053ad7acaddbf9e096745d142ec8d8b  clip-vit-b32-text.onnx
+924691ac288e54409236115652ad4aa250f48203de50a9e4722a6ecd48d6804a  bpe_simple_vocab_16e6.txt.gz
+```
+
+## Note on the CLIP pair
+
+`clip-vit-b32-image.onnx` and `clip-vit-b32-text.onnx` come from the **same
+export family** (both PyTorch 2.3.0 exports of `clip-ViT-B-32`) and therefore
+share one 512-d embedding space. Mixing an image encoder from one port with a
+text encoder from another puts the vectors in different spaces and makes cosine
+similarity meaningless. Treat the two as a single atomic unit.
+
+## Licensing
+
+Each file keeps its upstream license; see the table above and `LICENSES.md`.
+The `license: other` / `mixed-permissive` marker in the card metadata reflects
+that this repo aggregates MIT and Apache-2.0 files rather than being under one
+single license. No copyleft-licensed weights are hosted here.

arcface.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f3a6bc281e72f88862f5748b53be3d76b3b48f8f1ab1f4a537941bdc4e1b01da
+size 261036388

bpe_simple_vocab_16e6.txt.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:924691ac288e54409236115652ad4aa250f48203de50a9e4722a6ecd48d6804a
+size 1356917

clip-vit-b32-image.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c68d3d9a200ddd2a8c8a5510b576d4c94d1ae383bf8b36dd8c084f94e1fb4d63
+size 351686194

clip-vit-b32-text.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4dbe762b11e36488304471e439cde89da053ad7acaddbf9e096745d142ec8d8b
+size 254102519

yunet.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8f2383e4dd3cfbb4553ea8718107fc0423210dc964f9f4280604804ed2552fa4
+size 232589