Common recommendations on image rendering:
- For
batch_size_max(# images rendered in parallel), it is recommended to use ~20 for CPU and ~100 GPU. - For
img_size(dimension of rendered images), it is recommended to use ~64 for CPU, 512-1024 for GPU.
Below we show examples of sampling and interpolating on CPU with img_size 64 and batch_size_max 20.
The notebook sampling.ipynb provides code for random sampling from the generative model and rendering body images.
Details on sampling from different prior distributions can be found here.
The notebook interpolation.ipynb provides code for interpolating in the latent space and rendering sequences of body images.
The details on interpolation procedures can be found here.
