Results of DeepPrivacy

We use DeepPrivacy to anonymize videos.

DeepPrivacy

Link: https://github.com/hukkelas/DeepPrivacy Citation:

---
    @InProceedings{10.1007/978-3-030-33720-9_44,
    author="Hukkel{\aa}s, H{\aa}kon
    and Mester, Rudolf
    and Lindseth, Frank",
    title="DeepPrivacy: A Generative Adversarial Network for Face Anonymization",
    booktitle="Advances in Visual Computing",
    year="2019",
    publisher="Springer International Publishing",
    pages="565--578",
    isbn="978-3-030-33720-9"
    }
---

License: MIT

The faces are removed from original videos, and a generator is used to fill the blank part.

This ensures that the anonymized faces cannot be recovered. But, the new face is not restricted to a particular person. Hence, with time, the face in video changes in considerable quantity.

Also, the facial expression are totally lost. Hence, DeepPrivacy is almost like Hiding-faces.

Demo: https://drive.google.com/drive/folders/1obOc1rA0ydOH4zAI6BmbKy687s3oz41O?usp=sharing

IRB demos

Hiding face: https://drive.google.com/drive/folders/1Mrt–AqTC9ohYD457peYPZw0RvBHYSOK?usp=sharing

DeepPrivacy: https://drive.google.com/drive/folders/1obOc1rA0ydOH4zAI6BmbKy687s3oz41O?usp=sharing

What next?

Check several other DeepFake codes. Most of them are in tensorflow. But, this time, let’s directly test them on demo videos.