description |
---|
Notes and experiments on GANs. |
Tensorflow implementation of DCGANs [Generative Adversarial Networks]. Models trained on GCP with 1 x NVIDIA Tesla T4.
CIFAR-10 and CelebA DCGAN generated images. Approx. 60 epochs.
- Label Clipping/Label Smoothing
- BatchNorm layers in Generator model
- Change plotting and integrate Comet ML
- Streamlit.io
–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––
- MTCNN implementation for "face extraction"
- Create np.savez_compressed
- Tf-Gan model abstraction
- Latent Space SLERP interpolation