SFHQ-VirtualID-Bench Dataset Card

Summary

SFHQ-VirtualID-Bench is a synthetic, identity-conditioned face dataset of 750 synthetic identities, built as a benchmark for machine unlearning. Each identity is a deletion unit: all 90 aligned 224×224 crops of one identity share a single identity-level split (retain or forget), and 75 identities follow a sequential 15-step forgetting protocol. Every identity contributes both per-image image_subset values (train + holdout), so retention and forgetting-generalisation can be measured on genuinely held-out images (MUFAC-aligned evaluation).

The release ships two artifacts:

Artifact	Images	Per-identity	Purpose
dataset.csv (balanced)	67,500	90 (72 train + 18 holdout)	Primary unlearning benchmark; ships BOTH forget schedules as columns
dataset_imbalanced.csv (imbalanced)	36,064	34–100 (ratio-based)	Long-tail stress test — 5:1 popularity gradient (82:41:16 train)

This dataset is synthetic. Synthetic does not mean risk-free — residual likeness, demographic bias, and training-data memorisation are possible and must be acknowledged.

---

Composition

Property	Value
Identities	750
Resolution	224×224 aligned crops
Format	JPEG (uint8 BGR storage)
Images (balanced)	67,500 (90/id)
Images (imbalanced)	36,064 (34–100/id, ratio-based)
Splits	Retain: 675 identities · Forget: 75 identities
Forget protocol	15 steps, 75 identities — uniform 5/step (baseline) or seeded-Poisson batches (variant)
Holdout	18 per identity (max(min_holdout, round(imagesperidentity × holdout_frac)))
Imbalance gradient	5:1 — high 75 ids / medium 225 / low 450; train ratios {1.0, 0.50, 0.20} on an 82-train pool

The imbalanced artifact ships 36,064 rows (11 fewer than 75 × 100): 123 candidates failed the preprocess quality gate (blur/unreadable), capping 10 high-bin identity pools at 98–99 crops.

Labels

Column	Type	Values
identity_id	int	0–749
age_group	int	0=Young, 1=Adult, 2=Middle-Aged, 3=Senior
age	int	Raw InsightFace age estimate
gender	int	0/1 (InsightFace classifier)
split	string	retain, forget (identity-level)
image_subset	string	train, holdout (per-image; MUFAC-aligned)
forget_step	int	0–14 (forget only, uniform schedule), −1 (retain)
forget_step_poisson	int	0–14 (forget only, seeded-Poisson schedule), −1 (retain)
arcface_similarity	float	Cosine similarity to identity's mean embedding (confound control)
laplacian_variance	float	Sharpness score (quality confound control)
detection_confidence	float	Face detector confidence (alignment control)
popularity_bin	str	"high" / "medium" / "low" (imbalanced only)
images_per_identity	int	Actual per-identity count (imbalanced only)

• Schedule columns are balanced-only. The balanced artifact ships both forget_step (uniform baseline — step index == cumulative forgotten count) and forget_step_poisson (seeded arrival-model stress test, λ=5 rebalanced to 75; evaluate by cumulative count, not step index). The imbalanced artifact deliberately carries no schedule columns — its only experimental axis is the popularity gradient.
• Prompt metadata is stripped from Bench (pose/expression/lighting/ setting/camera ship only in the Raw release).
• Age labels are per-image, not per-identity. Each crop inherits the gender/age estimate taken on its parent 1024×1024 candidate, so one identity can span several age groups. This is intentional — do not collapse to a per-identity constant.
• Age-group labels are proxy estimates from the InsightFace classifier — not verified demographic attributes. The confound columns are provided so downstream MIA/fairness analyses can control for per-identity variation.

Split isolation invariant

Every identity_id maps to exactly one split (retain or forget), and every identity contributes both image_subset values (train + holdout). No identity's images appear in multiple splits. This invariant is enforced by the build step and validated by validate_release.py before publication.

Intended use

The 224×224 Bench is the primary machine-unlearning benchmark: aligned crops match the ImageNet training regime of downstream ResNet-18 classifiers, so pretrained features activate at full fidelity from epoch 1. The imbalanced variant adds a 5:1 popularity gradient for long-tail unlearning stress-testing, with a uniform 18-image holdout per identity so probe stability is comparable across all popularity tiers. Typical tasks: unlearning, membership-inference attacks, forgetting-generalisation, retention.

Method

Synthetic identities were generated with InstantID + Juggernaut-XL-v9 + ControlNet from CC0 synthetic SFHQ seed images (CLIP+KMeans-diverse seed selection); faces were detected, aligned, and cropped to 224×224, then quality-gated (sharpness + decodability) before split assignment. The arcface_similarity column is a confound control — the similarity floor was not applied as a filter in this release. Exact model revisions, generation configuration, and licence URLs are recorded in RELEASE_MANIFEST.json and THIRD_PARTY_NOTICES.md. No model weights are distributed.

Limitations

• Synthetic, not anonymous. Generated faces may retain unintended resemblance to real persons through the training data of the generators.
• Proxy labels. Age-group fields are model-estimated and carry classifier bias; they are not ground-truth demographics.
• Demographic bias. The SFHQ source may over-/under-represent certain appearances; no fairness correction has been applied.
• Artifacts. Diffusion artifacts (blur, distorted features) are most visible in 224×224 downscaled crops.
• Imbalance is synthetic. The popularity gradient is a deliberate down-sampling, not real-world distribution.

Privacy and release policy

The release does not include: raw SFHQ source images, seed-to-output linkage, rejected candidates, ArcFace embeddings or biometric templates, or model weights. Users must comply with the upstream licence terms of the generation models (see THIRD_PARTY_NOTICES.md). A takedown process for credible likeness complaints is available via the repository.

Licence and citation

• Code: MIT
• Images and metadata: Non-commercial research use only, with a prohibited-use clause (no biometric identification, surveillance, authentication, impersonation, or high-impact decisions about people). See LICENSE.

@dataset{sfhq_virtualid_bench,
  title     = {{SFHQ-VirtualID-Bench}: Synthetic Identity-Conditioned Aligned
               Face Crops for Machine Unlearning},
  author    = {Faiz Palwala},
  year      = {2026},
  version   = {1.0.0},
  doi       = {10.5281/zenodo.21877893},
  url       = {https://github.com/FaizPalwala/virtual-id-gen},
}

See CITATION.cff for the complete metadata file.

Acknowledgements

This work was undertaken on the Aire HPC system at the University of Leeds, UK.