Expand description
oxicuda-nerf — Neural Radiance Fields and neural rendering primitives for OxiCUDA.
Pure-Rust implementation of canonical NeRF algorithms with GPU PTX kernel generation.
§Architecture
oxicuda-nerf
├── camera/ — Pinhole camera model and ray generation
├── encoding/ — Positional encoding, Instant-NGP hash grid, Mip-NeRF IPE
├── error — NerfError / NerfResult
├── field/ — TensoRF, Instant-NGP hash field
├── handle — NerfHandle (SmVersion + LcgRng)
├── metrics/ — PSNR, MSE image quality metrics
├── network/ — NeRF MLP, TinyNeRF
├── ptx_kernels — GPU PTX kernel strings (7 kernels × 6 SM versions)
└── rendering/ — Ray, sampling, volume rendering, occupancy gridModules§
- camera
- Camera models.
- encoding
- Encoding modules for NeRF input features.
- error
- Error types for
oxicuda-nerf. - field
- Neural field representations.
- handle
- Session handle for
oxicuda-nerf. - metrics
- Image quality metrics for NeRF evaluation.
- network
- Neural network modules for NeRF.
- prelude
- Convenience re-exports for common NeRF types and functions.
- ptx_
kernels - PTX GPU kernel sources for NeRF and neural rendering operations.
- rendering
- Rendering modules: ray generation, sampling, volume rendering, occupancy.