Lower live Leap packed stereo to depth socket

This commit is contained in:
Meta Vens 2026-05-30 21:54:58 +01:00
commit 2c197cb3ae
9 changed files with 247 additions and 36 deletions

View file

@ -408,20 +408,22 @@ resources, and explicit fences. CUDA, TensorRT, and monocular depth demos remain
research references only.
The first live contract exists: `MimirStereoDepthConfigurations` names the
libSGM-provenance D3D12 SGM profile, and
`MimirFensalirFieldLowering.BuildStereoDepthCandidateFrame` lowers a synthetic
rectified stereo result by referencing caller-declared shader-readable
left/right input textures, then emitting a GPU-resident compute-writable
`MimirFensalirFieldLowering.BuildStereoDepthCandidateFrame` lowers caller-
declared rectified stereo input textures into a GPU-resident compute-writable
disparity SurfacePage, confidence texture, Height FieldEvidence claim, and
`AquariumFieldStereoDepthLowering` sidecar. The lowering sidecar is the dispatch contract: it binds the
`AquariumFieldStereoDepthLowering` sidecar. The live
`BuildLeapPackedStereoDepthCandidateFrame` path now recognizes a live
`LeapStereoIr` rolling video window, reuses its declared packed R8G8 texture as
both left and right resource identity, names the left/right observations as
packed R/G lanes, and emits the same depth lowering automatically from the
runtime frame. The lowering sidecar is the dispatch contract: it binds the
libSGM-provenance profile, calibration id, camera pair, inputs, disparity
output, min disparity, disparity levels, aggregation paths, census radius,
P1/P2 smoothness penalties, and depth range. The current D3D12 SGM profile
publishes min disparity 0, 128 disparities, four paths, census radius 2, P1 8,
and P2 96. The kernel itself is still future work. Fensalir's D3D12 renderer now counts stereo-depth
lowerings as dispatch-ready only when the Height claim is planned and the
left/right input textures plus UAV disparity `SurfacePage` resolve through its
field resource registry; it logs that readiness separately from the absent
kernel.
and P2 96. Fensalir now has a first packed-Leap D3D12 kernel that writes R16F
disparity through a crude SAD/block-match pass. This is intentionally a small
kernel-shaped proof, not full libSGM-style SGM or calibrated metric depth yet.
Fensalir must not be treated as a traditional rendering pipeline. Mimir does
not ask it to "draw a thing" and hope post-processing makes the result true.