I am developing Deep Video Analytics [1] which incorporates FaceNet [2] and MTCNN detector. 1. What makes you think that governments would openly release their

I am developing Deep Video Analytics [1] which incorporates FaceNet [2] and MTCNN detector. 1. What makes you think that governments would openly release their recognition models? 2. Do you have any hard numbers? 3. What's the difference between manipulating an image vs. just blurring/blacking it out, since the latter essentially gives you complete anonymity. In my experience all Adversarial techniques fall apart when another algorithm is added to the mix (including the recent ones by OpenAI [3]). As a result I am frankly distrustful of any claim about protecting identity. The State of Art in Face Recognition moves extremely fast, and these are just publicly available / permissively licensed models, and not what Government agencies around the world are experimenting on with their billion face datasets. [4] [1] http://www.deepvideoanalytics.com/ http://www.deepvideoanalytics.com/ [2] https://github.com/davidsandberg/facenet/ https://github.com/davidsandberg/facenet/ [3] https://www.dropbox.com/s/oshcos8wzvqtgow/Screenshot%202017-07-25%2013.44.23.png?dl=0 https://www.dropbox.com/s/oshcos8wzvqtgow/Screenshot%202017-... [4] https://github.com/wy1iu/sphereface https://github.com/wy1iu/sphereface