Drop-in, idempotent speed patches for PGSR surface-reconstruction training — removes per-iteration PCIe transfers, redundant ones-kernel convolutions, and GPU-CPU logging syncs (plus a real EMA-logging bug) with CPU-proven numerical parity: same geometry, lower GPU bill.
performance computer-vision deep-learning cuda pytorch surface-reconstruction photogrammetry nerf 3d-reconstruction gpu-optimization mesh-reconstruction gaussian-splatting training-optimization 3d-gaussian-splatting radiance-fields pgsr
-
Updated
Jul 26, 2026 - Python