Module camera 

Camera-pose conditioning MLP.

Embeds the flattened 4×3 extrinsics matrix (12 floats) into the same dimension as the U-Net time embedding so it can be added to the timestep conditioning signal.

Structs

CameraEmbedding

MLP that lifts a flat camera-pose vector to the time-embedding dimension.

TimestepEmbedding

Timestep-embedding MLP (projects sinusoidal embeddings to a wider space).

Functions

timestep_embedding

Build sinusoidal timestep embeddings (same as Stable Diffusion).