Notation

Symbol	Meaning
\(\mathbf{x} \in \mathbb{R}^{H \times W \times C}\)	An image with height \(H\), width \(W\), channels \(C\)
\(\mathbf{x} * \mathbf{k}\)	Convolution of image \(\mathbf{x}\) with kernel \(\mathbf{k}\)
\(f_\theta\)	A network with parameters \(\theta\)
\(\mathcal{L}\)	A loss function
\(\mathbf{q}, \mathbf{k}, \mathbf{v}\)	Query, key, value vectors in attention
\(\mathbf{z}\)	A latent / embedding vector
\(\mathbf{x}_t\)	The image at step \(t\) of a noising process (so \(\mathbf{x}_0\) is clean data)
\(q(\mathbf{x}_t \mid \mathbf{x}_0)\)	The forward (noising) process
\(p_\theta(\mathbf{x}_{t-1} \mid \mathbf{x}_t)\)	The learned reverse (denoising) process
\(\epsilon_\theta(\mathbf{x}_t, t)\)	The noise-prediction network
\(\mathbf{K}, [\mathbf{R} \mid \mathbf{t}]\)	Camera intrinsics; extrinsics (rotation, translation)

Sets are calligraphic (\(\mathcal{D}\) for a dataset), matrices bold uppercase, vectors bold lowercase, scalars italic. Conventions are introduced properly where they first matter; this page is the lookup table.