Algorithm 1: Plane-sweep multi-image matching executed at a specific pyramid level l of the proposed hierarchical processing scheme.
	Data:a calibrated image bundle ${\mathsf{\Omega }}^l$ at the pyramid level l, a set of planes $\mathsf{\Pi }$ with a normal vector $\mathrm{n}$ and varying distances $\delta$ as well as a local depth sampling range ${\Gamma }_{\mathrm{p}}^l=[d_{\mathrm{p},\min }^l,d_{\mathrm{p},\max }^l]$.
	Result: three-dimensional cost volume $\mathcal{S}$, holding the pixel-wise matching score for each pixel ${\mathrm{p}}^{\mathrm{ref}}\in {\mathcal{I}}_{\mathrm{ref}}^l$ and plane $\mathsf{\Pi }$.
1	determine bounding planes ${\mathsf{\Pi }}_{\min }$ and ${\mathsf{\Pi }}_{\max }$ located at ${\delta }_{\min }$ and ${\delta }_{\max }$, so that the local depth range ${\Gamma }_{\mathrm{p}}^l$ is completely sampled (see Section 2.1.2).
2	foreach pixel ${\mathrm{p}}^{\mathrm{ref}}\in {\mathcal{I}}_{\mathrm{ref}}^l$ and distance $\delta \in \left[{\delta }_{\min },{\delta }_{\max }\right]$ do $\phantom{(}$$\phantom{(}$
3		Configure scene plane $\mathsf{\Pi }=(\mathrm{n},\delta )$.
4		Determine pixels ${\mathrm{p}}^k$ in all matching images ${\mathcal{I}}_k^l\in {\mathsf{\Omega }}^l\setminus {\mathcal{I}}_{\mathrm{ref}}^l$:
		$${\mathrm{p}}^k=\mathrm{H}\left(\mathsf{\Pi },{\mathrm{P}}_{\mathrm{ref}}^l,{\mathrm{P}}_k^l\right)·{\mathrm{p}}^{\mathrm{ref}}.$$
5		Warp local image patches ${\mathcal{P}}_k^l\in {\mathcal{I}}_k^l$ around ${\mathrm{p}}^k$, with the same size as the support region of the matching cost function $C(·)$, into ${\mathcal{I}}_{\mathrm{ref}}^l$:
		$${\tilde{\mathcal{P}}}_k^l=\mathrm{H}{\left(\mathsf{\Pi },{\mathrm{P}}_{\mathrm{ref}}^l,{\mathrm{P}}_k^l\right)}^{-1}·{\mathcal{P}}_k^l.$$
6		Compute the matching cost $s\left(\mathrm{p},\mathsf{\Pi }\right)$ between reference patch ${\mathcal{P}}_{\mathrm{ref}}^l\in {\mathcal{I}}_{\mathrm{ref}}^l$ and ${\tilde{\mathcal{P}}}_k^l$ for left and right subset of cameras separately:
		$$s^{\mathrm{L}}\left(\mathrm{p},\mathsf{\Pi }\right)=\sum _{k<\mathrm{ref}}C\left({\mathcal{P}}_{\mathrm{ref}}^l,{\tilde{\mathcal{P}}}_k^l\right),$$
		$$s^{\mathrm{R}}\left(\mathrm{p},\mathsf{\Pi }\right)=\sum _{k>\mathrm{ref}}C\left({\mathcal{P}}_{\mathrm{ref}}^l,{\tilde{\mathcal{P}}}_k^l\right).$$
7		Store the minimum of left and right matching cost (accounting for occlusions as described by Kang et al. [50]) into three-dimensional cost volume $\mathcal{S}$:
		$${\mathcal{S}}^l\left(\mathrm{p},\mathsf{\Pi }\right)=min\{s^{\mathrm{L}}\left(\mathrm{p},\mathsf{\Pi }\right),s^{\mathrm{R}}\left(\mathrm{p},\mathsf{\Pi }\right)\}.$$
8	end $\phantom{(}$