AdaFace IR-101 checkpoint trained on WebFace12M and exported for face-similarity research.
This repository contains immutable model artifacts used by
Facetorch. Use the packaged Facetorch manifest to select a revision and artifact; do not treat mutable
main or older unlisted files as a release contract.
Preprocessing: Resize to 112 by 112, convert RGB to BGR, then normalize with mean and standard deviation [0.5, 0.5, 0.5].
Facetorch v1 supports the Torch 2.6 and 2.11 cohort files listed in its manifest. The legacy TorchScript object is CPU-only and requires the explicit legacy opt-in. Files from unsupported cohorts are not part of the v1 release contract.
1import torch
2from huggingface_hub import hf_hub_download
3from facetorch.artifacts import get_model_manifest
4
5MODEL_ID = "verify-adaface"
6device = "cuda" if torch.cuda.is_available() else "cpu"
7artifact = get_model_manifest().candidates(
8 MODEL_ID,
9 torch_version=torch.__version__,
10 device=device,
11 allow_legacy_models=False,
12)[0]
13path = hf_hub_download(
14 repo_id=artifact.repo_id,
15 revision=artifact.revision,
16 filename=artifact.filename,
17)
18model = torch.export.load(path).module().to(device).eval()
19example = torch.randn(1, 3, 112, 112, device=device)
20with torch.inference_mode():
21 output = model(example)
The random tensor above is only a loading smoke test. Use Facetorch's documented preprocessing for meaningful inference.
Result: Seeded batch-one and batch-two cases matched with maximum embedding absolute error 1.05e-6 and maximum norm absolute error 2.01e-5.
The repository owner approved the mapping and redistribution record on 2026-08-23. Under the recorded policy, an author-published checkpoint in a permissively licensed repository with no separate checkpoint terms uses that repository license. MIT and Apache-2.0 have not been converted or treated as interchangeable. See
LICENSE,
THIRD_PARTY_NOTICES.md, and Facetorch's
facetorch/models/governance.json.
