Paper #1 - Deep Learning for Deepfakes Creation and Detection: A Survey
Key Notes
- Deep Fake Models
How it works
- A deepfake creation model using two encoder-decoder pairs
- Two networks use the same encoder but different decoders for training process
- An image of face A is encoded with the common encoder and decoded with decoder B to create a deepfake (bottom).
Deep Fake Detection Techniques
- Eye blinking
- Inconsistency between frames
- Face warping artifacts, Inconsistency of surrounding areas
- Head poses
- Exploit facial texture differences
- Emotion audiovisual affective cues
Paper #2 - Video and audio deepfakes detection using Deep Learning
Key Notes
To summarize, given an image of a face, the FWA facial warping artifacts (FWA) captures the face features using dlib CNN face detector then it blurs the captured face feature and affine warps it back to the original face image. This procedure emulates deepfake faces and it is much simpler to generate compared to using autoencoders
We will be feeding in both the real and fake faces into the neural network and this may cause confusions during the training phase
Projects link
Keep Exploring!!!
No comments:
Post a Comment