Hi DF40 team! 👋
Congratulations on NeurIPS 2024 acceptance — DF40 looks like an amazing next-gen dataset with 40 different deepfake techniques!
I wanted to reach out about SiliconSignature (repo | demo), our open-source hardware-bound image authentication system.
Why this fits DF40:
Your dataset has 40 forgery methods. What if the original images had hardware-bound watermarks that survive all 40? This would let you:
- Verify ground-truth authenticity of source images
- Create a "watermark survival benchmark" — which forgeries destroy the signature?
- Add a 41st detection dimension: "watermark present = original"
What SiliconSignature offers:
- 🔏 Real ASIC proof-of-work (Antminer S9 BM1387)
- 🛡️ Reed-Solomon survives 40% pixel destruction
- 🎨 Invisible LSB watermark — doesn't affect visual quality
- 💻 Pure Python, MIT license, zero dependencies
Collaboration idea:
We could watermark your original source videos and measure which of the 40 techniques destroy or preserve the signature. This could be a novel evaluation metric in your paper.
Would love to collaborate! Let me know if you're interested.
Best,
Francisco Angulo de Lafuente (@Agnuxo1) — P2PCLAW Research
Hi DF40 team! 👋
Congratulations on NeurIPS 2024 acceptance — DF40 looks like an amazing next-gen dataset with 40 different deepfake techniques!
I wanted to reach out about SiliconSignature (repo | demo), our open-source hardware-bound image authentication system.
Why this fits DF40:
Your dataset has 40 forgery methods. What if the original images had hardware-bound watermarks that survive all 40? This would let you:
What SiliconSignature offers:
Collaboration idea:
We could watermark your original source videos and measure which of the 40 techniques destroy or preserve the signature. This could be a novel evaluation metric in your paper.
Would love to collaborate! Let me know if you're interested.
Best,
Francisco Angulo de Lafuente (@Agnuxo1) — P2PCLAW Research