Generative Models
BigGAN-deep
Introduced by Brock et al. in Large Scale GAN Training for High Fidelity Natural Image SynthesisBigGAN-deep is a deeper version (4x) of BigGAN. The main difference is a slightly differently designed residual block. Here the $z$ vector is concatenated with the conditional vector without splitting it into chunks. It is also based on residual blocks with bottlenecks. BigGAN-deep uses a different strategy than BigGAN aimed at preserving identity throughout the skip connections. In G, where the number of channels needs to be reduced, BigGAN-deep simply retains the first group of channels and drop the rest to produce the required number of channels. In D, where the number of channels should be increased, BigGAN-deep passes the input channels unperturbed, and concatenates them with the remaining channels produced by a 1 × 1 convolution. As far as the network configuration is concerned, the discriminator is an exact reflection of the generator.
There are two blocks at each resolution (BigGAN uses one), and as a result BigGAN-deep is four times deeper than BigGAN. Despite their increased depth, the BigGAN-deep models have significantly fewer parameters mainly due to the bottleneck structure of their residual blocks.
Source: Large Scale GAN Training for High Fidelity Natural Image Synthesis
Papers
| Paper | Code | Results | Date | Stars |
|---|
Tasks
| Task | Papers | Share |
|---|---|---|
| Image Generation | 4 | 18.18% |
| Conditional Image Generation | 3 | 13.64% |
| Bias Detection | 2 | 9.09% |
| Clustering | 2 | 9.09% |
| Density Estimation | 1 | 4.55% |
| Metric Learning | 1 | 4.55% |
| Relational Reasoning | 1 | 4.55% |
| Self-Supervised Learning | 1 | 4.55% |
| Fairness | 1 | 4.55% |
Bottleneck Residual Block
Conditional Batch Normalization
Early Stopping
GAN Hinge Loss
Linear Layer
Non-Local Block
Off-Diagonal Orthogonal Regularization
Projection Discriminator
SAGAN
Spectral Normalization
Truncation Trick
TTUR
