[Zador-Pataki]

Motivation

Use one prepared-data layout across datasets so loading, evaluation and playback share the same recording and camera conventions.

Changes

• Store data under <dataset>/<scene>/<recording>/<variant>/, with images/camN/ and a reference model containing only calibration and poses.
• Adapt all supported datasets while preserving benchmark selections and report grouping.
• Resolve input loading, cache identity and playback from explicit recording and variant selections; share preparation helpers across datasets.
• Share cam0 trajectory evaluation between monocular and rig inputs.
• Configure panorama view directions, FOV and aspect ratio at runtime, rendering directly at the tracking resolution. Support stitched video input.