Upstream training splits; known explicitly identified test/eval rows excluded. Cross-dataset benchmark overlap is not guaranteed. Published as a raw, manually gated release; source annotation caveats remain.

Task views reuse original image archives. 2D coordinates are normalized 0–1000. Native 3D sidecars preserve original camera-space XYZ and camera calibration; they are not normalized to 0–1000. Within each query targets are sorted left-to-right then top-to-bottom.

HF preview configs contain 10 examples per view, not the complete training split. Full training uses metadataset.yaml and examples/read_energon.py (Energon 7.4).

Source	View	Records	Targets	Examples
ca1m	boxes	1,505,783	18,107,210	10 annotated examples
omni3d	boxes	220,705	2,596,296	10 annotated examples

What is not claimed

No guarantee of cross-dataset benchmark disjointness, global image uniqueness, semantic annotation correctness, or redistribution/license clearance. Shared box/point views are alternate tasks, not distinct images. Source-specific caveats remain applicable.

The package preserves upstream query text; metadata cleanup does not alter images or embedded image metadata.

Optional native 3D annotations

CA-1M native boxes and camera metadata are provided as optional annotation sidecars. Existing 2D views, image archives and metadataset.yaml are unchanged. These are not a 3D training codec.

Downloads last month: 21

Total file size:

893 GB