Encoder-Decoder
Definition
A neural architecture where an encoder compresses an input into a compact latent representation and a decoder reconstructs an output image from it. Face-swap tools train separate decoders for each identity on a shared encoder, allowing one face's expression to drive another's appearance.
- Encoder role
- Compresses input to latent representation
- Decoder role
- Reconstructs an output image
- Deepfake use
- Shared encoder, per-identity decoders
Common questions
How does a shared encoder enable a face swap?+
The encoder learns identity-agnostic features such as expression, pose, and lighting common to all training faces, while each identity's dedicated decoder renders those features back using that specific person's appearance.
Does this architecture only appear in deepfakes?+
No. Encoder-decoder architectures are a general neural network pattern used for image compression, translation, and denoising as well, not exclusive to face-swap generation.
Related terms
- Diffusion Model
- A generative neural network architecture (Ho et al., 2020; Stable Diffusion, Rombach et al., 2022) that learns to reverse a noise-addition process...
- GAN
- Generative Adversarial Network. A framework with two neural networks, a generator that creates synthetic data and a discriminator that tries to distinguish...
- Latent Space
- The compressed, lower-dimensional representation of data learned by a neural network's internal layers. Generative models sample from or navigate this space to...
- NeRF (Neural Radiance Field)
- A neural representation that encodes a 3-D scene as a continuous volumetric function, allowing novel viewpoints to be rendered. In talking-head systems,...
- Voice Conversion
- A signal-processing or deep-learning technique that transforms the vocal characteristics of a source speaker's utterance to match a target speaker, while preserving...