| Version | Supported |
|---|---|
| Latest (main) | ✔ |
| Older commits | ✗ |
Do NOT open a public GitHub issue for security vulnerabilities.
Report privately via GitHub Security Advisories:
- Go to the repository → Security tab → Report a vulnerability
- Provide a detailed description, reproduction steps, and potential impact
- You will receive a response within 72 hours
This codebase processes human facial images. By using this repository you agree to:
- Consent — Only process images of individuals who have given explicit informed consent
- No mass surveillance — Do not use for bulk de-anonymization, surveillance, or identity tracking
- No deception — Do not use outputs to deceive, impersonate, or defraud individuals
- No biometric databases — Do not use model outputs to build unauthorized biometric databases
- Authorized contexts only — Permitted uses: academic research, authorized forensics, security testing with consent
Violation of this policy also violates the CC BY-NC-ND 4.0 license under which this code is released.
| Risk | Description |
|---|---|
| Privacy | Model can reveal the appearance of individuals without makeup — use only with consent |
| Bias | Trained on FFHQ-Makeup; may perform differently across skin tones and demographic groups |
| Misuse | Could be used to "unmask" individuals who prefer their cosmetic presentation |
| No guarantees | Outputs are model predictions, not ground truth — do not use for legal identification |
Pretrained weights are stored in a private Hugging Face repository (priyadip/itdf-model).
Access requires explicit approval. Approved access does not grant redistribution rights.
Keep dependencies up to date. Known critical packages:
pip install --upgrade torch torchvision timm PillowRun pip audit periodically to check for vulnerabilities in dependencies.