Deep Fakes 2.0 - How Neural Networks are Changing our World
summary
A working real-time face transfer prototype maps live facial expressions onto other people (examples include Barack Obama and Angela Merkel) by combining Keras-trained autoencoders, FaceNet-style landmarking, OpenCV/TensorFlow face detection, and MobileNet+UNet segmentation; training uses two Nvidia 1080 Ti GPUs and ~5 hours (DeepFaceLab baseline) plus a 35,000-face segmented dataset and 17–18 decoders before switching to a GAN-style discriminator to improve realism — the pipeline runs at interactive frame rates.
AI-written summary · about
chapters (16)
- 0:00Introduction and speakers
- 1:25What are deep fakes
- 3:08Obama lip sync example
- 8:41Face swap vs deep fakes
- 11:38Deep fake definition
- 14:30Deep Face Lab overview
- 20:13Autoencoder training
- 26:17Motivation for real-time
- 28:24Real-time implementation
- 33:06Face segmentation
- 36:44Mask blending
- 37:05Generative adversarial networks
- 45:28Live demo of real-time
- 47:17Fields of application
- 49:52Legal and ethical concerns
- 51:17Q&A session
related talks
58mPushing Deepfakes to the Limit-Fake video calls with AI
51mFaceswap: Developing a platform for practical Deepfakes
51mPlaying Offense and Defense with Deepfakes
38mDIY Deep Fakes
01h 03mIt's All About AI
28mChallenges & future trends of Deep Learning
03mBuilding artificial people: Endless possibilities and the dark side
46mHow I built the world’s most efficient deepfake detector with $100