DEEP CONVOLUTIONAL FACE SYNTHESIZER
Deep learning generative modeling system using a conditional deep convolutional GAN to synthesize and edit faces by attributes (gender, hair, hat, eyeglasses, smiling), with a small desktop app.
- Framework
- TensorFlow
DCGAN SYNTHESIS GRID, FEMALE LATENT SAMPLES
DCGAN & CGAN FACIAL GENERATION & ATTRIBUTE MORPHING
Behind the project.
A conditional deep convolutional GAN implemented in TensorFlow: a generator and discriminator trained with attribute condition vectors, with OpenCV Haar-cascade face detection and alignment in the data pipeline and latent-space interpolation between generated identities. A Kivy desktop app provides random-face, sketch and face-manipulation screens. Repo examples: 128x128 faces at 150 epochs (female, male, hat, bald, eyeglasses, blond, anime).
Project media.
3 photos & media · Tap to expand
Under the hood.
- Framework
- TensorFlow
- Architecture
- Conditional DCGAN (BatchNorm + LeakyReLU)
- Conditioning
- Attribute Vectors (Bald, Blond, Eyeglasses, Hat, Smiling, ...)
- Preprocessing
- OpenCV Haar-Cascade Face Detection & Alignment
- Latent Dimension
- 100
- App
- Kivy Desktop UI
What came out of it.
- Attribute-conditioned face generation and morphing.
- Haar-cascade face alignment pipeline.
- Latent-space interpolation between identities.