CalibRaw third-party artifact mirror
This Hugging Face repository is the stable artifact mirror for CalibRaw. It preserves the model and runtime payloads CalibRaw depends on so application downloads can be pinned to an immutable repository revision.
These artifacts were not authored by the maintainer of this mirror. They
remain the work of their respective upstream projects, and every artifact
retains its original license. This repository is a multi-license collection;
there is no single license that replaces the upstream terms. See
licenses/README.md before redistributing an artifact.
Repository layout
models/
briefnet/ BiRefNet ONNX checkpoints (historical directory spelling)
lama/ Big-LaMa inpainting ONNX model
rawnind/ RawNIND .dtmodel package with two embedded ONNX models
sam2/ SAM 2.1 Hiera Tiny ONNX encoder and decoder
onnxruntime/ Official ONNX Runtime release archives by OS and CPU
licenses/ Upstream licenses, third-party notices, and artifact map
ARTIFACTS.md Per-file provenance, version, digest, and source audit
SHA256SUMS.txt SHA-256 digests for every distributed model/runtime payload
The models/briefnet/ spelling and all existing artifact filenames are kept
for compatibility and mirror stability.
Included models
| Component | Artifact/version | Purpose | Platform / architecture | License |
|---|---|---|---|---|
| BiRefNet | Release v1: General-Lite epoch 232, Lite-2K epoch 232, HR epoch 130 |
Subject/background segmentation | Portable ONNX; execution-provider dependent | MIT |
| SAM 2.1 | Hiera Tiny encoder + decoder, conversion revision aa116690… |
Prompted object segmentation | Portable ONNX; execution-provider dependent | Apache-2.0 |
| Big-LaMa | Places2 FP32, conversion revision a3ee2fca… |
Image inpainting | Portable ONNX; execution-provider dependent | Apache-2.0 |
| RawNIND | darktable-ai release 5.6.0 .dtmodel package, Bayer + Linear Rec.2020 UtNet2 |
RAW denoising/demosaicing | Package containing portable ONNX models; execution-provider dependent | GPL-3.0 |
Model files do not target a particular desktop OS or CPU instruction set. Runtime compatibility still depends on the ONNX Runtime build, supported operators, and the selected execution provider.
Included ONNX Runtime archives
These are unmodified official CPU release archives from microsoft/onnxruntime. Their local digests match the SHA-256 digests published in GitHub's release asset metadata.
| OS | CPU architecture | ONNX Runtime | Upstream archive |
|---|---|---|---|
| Linux | ARM64 / AArch64 | 1.29.0 | onnxruntime-linux-aarch64-1.29.0.tgz |
| Linux | ARM64 / AArch64 | 1.30.0 | onnxruntime-linux-aarch64-1.30.0.tgz |
| Linux | x86-64 | 1.29.0 | onnxruntime-linux-x64-1.29.0.tgz |
| Linux | x86-64 | 1.30.0 | onnxruntime-linux-x64-1.30.0.tgz |
| macOS | Apple Silicon / ARM64 | 1.29.0 | onnxruntime-osx-arm64-1.29.0.tgz |
| macOS | Apple Silicon / ARM64 | 1.30.0 | onnxruntime-osx-arm64-1.30.0.tgz |
| macOS | Intel x86-64 | 1.23.2 | onnxruntime-osx-x86_64-1.23.2.tgz |
| Windows | ARM64 | 1.29.0 | onnxruntime-win-arm64-1.29.0.zip |
| Windows | ARM64 | 1.30.0 | onnxruntime-win-arm64-1.30.0.zip |
| Windows | x86-64 | 1.29.0 | onnxruntime-win-x64-1.29.0.zip |
| Windows | x86-64 | 1.30.0 | onnxruntime-win-x64-1.30.0.zip |
Both 1.29.0 and 1.30.0 archives are retained for Linux, macOS ARM64, and
Windows. The macOS x86-64 archive is version 1.23.2. Each archive contains
its own LICENSE and
ThirdPartyNotices.txt; repository-level copies are under licenses/.
Verify downloads
From the repository root, verify every distributed model and runtime archive:
sha256sum --check SHA256SUMS.txt
On macOS, where GNU sha256sum may not be installed, verify an individual
file with:
shasum -a 256 path/to/artifact
Compare the result with SHA256SUMS.txt. A mismatch means
the file must not be used.
Immutable download pins
Production CalibRaw builds should download from an immutable Hugging Face
commit, not the mutable /main/ branch. Use:
https://huggingface.co/Duecki/CalibRaw-Artifacts/resolve/<commit-sha>/<artifact-path>
For example, replace <commit-sha> with the full 40-character commit ID that
was reviewed and whose files match the pinned CalibRaw hashes. Moving a tag or
branch is not an equivalent integrity guarantee.
Upstream sources and attribution
- BiRefNet release v1 by ZhengPeng and contributors.
- SAM 2 by Meta and contributors; ONNX conversion from the immutable akiyamanx revision.
- LaMa by Samsung Research and contributors; immutable Carve LaMa-ONNX revision.
- darktable-ai RawNIND release 5.6.0, based on the Raw Natural Image Noise Dataset work identified in its model card.
- ONNX Runtime releases by Microsoft and contributors.
Exact source filenames, immutable revisions, hashes, and the comparison with
CalibRaw's expected downloads are in ARTIFACTS.md.
Disclaimer
This mirror is provided for reproducible CalibRaw distribution. It does not claim authorship, ownership, sponsorship, or endorsement of the mirrored projects. Consult the upstream licenses and notices for your obligations.