You can deploy this model and start earning money today!

tomas-gajarsky/
facetorch-verify-magface

Name: facetorch-verify-magface
Availability: InStock

Views

No views yet

MagFace plus UNPG ResNet-100 verifier

MagFace ResNet-100 trained with Unified Negative Pair Generation 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.

Contract

Field	Value
Model ID	`verify-magface`
Architecture	MagFace IResNet-100 with UNPG training
Input	112 by 112 RGB face crop
Output	One 512-dimensional embedding per face.
Dynamic shapes	Batch dimension 1 through 64.
Weights license	Apache-2.0

Preprocessing: Resize to 112 by 112 and apply ImageNet mean [0.485, 0.456, 0.406] and standard deviation [0.229, 0.224, 0.225], matching the Facetorch contract.

Release artifacts

File	Format	Runtime	Devices	SHA-256
`model-torch2.6.pt2`	pt2	>=2.6, <2.7	cpu, cuda	`d6977f226493a55f72a54c57906998b8b814721a7560030e07c216fb9c9a6de1`
`model-torch2.11.pt2`	pt2	>=2.11, <2.12	cpu, cuda	`3298c38e5cb26aa00bac8128627f365dd47f57345a034e8b33f3742cde63f6eb`
`model.pt`	torchscript	>=2.6, <2.12	cpu	`0c54aab654369235b25c4df96214cf9fda23c8535a48eb5c9a0c8a03c79601dc`

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.

Loading the manifest-selected artifact

python

import torch
from huggingface_hub import hf_hub_download
from facetorch.artifacts import get_model_manifest

MODEL_ID = "verify-magface"
device = "cuda" if torch.cuda.is_available() else "cpu"
artifact = get_model_manifest().candidates(
    MODEL_ID,
    torch_version=torch.__version__,
    device=device,
    allow_legacy_models=False,
)[0]
path = hf_hub_download(
    repo_id=artifact.repo_id,
    revision=artifact.revision,
    filename=artifact.filename,
)
model = torch.export.load(path).module().to(device).eval()
example = torch.randn(1, 3, 112, 112, device=device)
with torch.inference_mode():
    output = model(example)

The random tensor above is only a loading smoke test. Use Facetorch's documented preprocessing for meaningful inference.

Provenance

Upstream	Immutable revision	Role	License
https://github.com/junuke/UNPG	`a6f9c1731a68fc035eb8fe8198f5a5c643825a5b`	checkpoint publisher and UNPG training source	Apache-2.0
https://github.com/IrvingMeng/MagFace	`99bae614ac2643b9694bf18e0c5645272ff6acfa`	MagFace architecture and loss source	Apache-2.0

Upstream checkpoint	SHA-256	Source
`MagFace+UNPG ResNet-100 checkpoint`	`Unavailable from expired source`	publisher location

Mapping method: founder_attested_chain_of_custody_corroborated_by_repository_history_and_negative_control.

Result: The founder confirmed acquisition from the author repository; Facetorch commit afff284236bdbf4cdd6271944791d085baca963c introduced the model as MagFace+UNPG on 2022-08-31; the immutable TorchScript archive identifies models.iresnet.IResNet; and all 925 comparable tensors differ from the authors' plain MagFace R100 checkpoint, excluding accidental substitution with that baseline.

Evidence limitation: No current byte-for-byte comparison to the expired author download is possible; this chain-of-custody method is intentionally disclosed on the model card.

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.

Papers

Intended use

Consent-based research on face-similarity embeddings with independently calibrated thresholds.

Limitations and responsible use

Similarity is not proof of identity and false matches or non-matches can cause serious harm.
The original author checkpoint URL is no longer downloadable, so provenance relies on the disclosed founder-attested chain of custody rather than current tensor equality.
The checkpoint license does not grant rights to MS1MV2 or other training datasets.
The artifact license does not itself license training datasets, input data, or a deployment's processing of personal data.
Do not use model output as the sole basis for consequential decisions about a person.

tomas-gajarsky/facetorch-verify-magface