render frame(s) of mesh using pytorch3d

Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).

Transformer block for image-like data. First, project the input (aka embedding) and reshape to b, t, d. Then apply standard transform

Modified original version to filter out-of-box samples as TensoRF does. Reference: TensoRF: https://github.com/apchenstu/TensoRF/blo

Triplane decoder that gives RGB and sigma values from sampled features. Using ReLU here instead of Softplus in the original implementation.

using Modified Block, see below

Convenience class that behaves like a dict but allows access with the attribute syntax.

Uses the dinov2 image encoder with camera modulation. Not actually frozen... If you want that set cond_stage_trainable=False in cfg

The full UNet model with attention and timestep embedding. :param in_channels: channels in the input Tensor. :param model_channels: base

Redirect stderr to stdout, optionally print stdout to a file, and optionally force flushing on both stdout and the file.

2D image to patch embedding: (B,C,H,W) -> (B,N,D) Args: img_size: Image size. patch_size: Patch token size. in_chans

Synthesizer that renders a triplane volume with planes and a camera. Reference: EG3D: https://github.com/NVlabs/eg3d/blob/main/eg3d/