API reference

The full, generated reference for every header lives at API reference (Doxygen). This page maps what you want to do to the right header and class.

Open the Doxygen reference Class list File list

1. Headers
2. Sparse tensors
3. Layers
4. Training
5. Adaptive voxelization
6. Gaussian splats
7. Source

---

Headers

Most programs need one include:

#include <riemann.hpp>             // sparse tensors, layers, training, adaptive, Gaussians*
#include <octree/adaptive.hpp>     // adaptive voxelization only

* Gaussian headers are included when the library was built with RIEMANN_ENABLE_GAUSSIAN (on in all release packages). The installed CMake target defines it for you.

Canonical headers live in engine/, octree/ and splat/. The headers in layers/, helpers/, dynamic/, normal/, core/ and learning/, and most root-level headers, only forward to them and remain for compatibility.

Namespace	Contents
riemann	Sparse tensors, coordinate managers, layers, functional ops, batching
riemann::training	Optimizers, batches, training and evaluation loops
riemann::adaptive	Adaptive (octree) voxelization and convolution
riemann::gaussian	3D Gaussian splats: model, rendering, training, PLY

Sparse tensors

Type / function	Header	Guide
SparseTensor	engine/tensor/sparse_tensor.hpp	Sparse tensors
TensorField	engine/tensor/field.hpp	TensorField
CoordinateManager	engine/grid/space.hpp	Core concepts
SparseQuantizeResult, sparse_quantize, batched_coordinates, sparse_collate	helpers/batching.hpp, helpers/discretization.hpp	Sparse tensors
cat, sum, mean, var, to_sparse, to_sparse_all	engine/nn/operations.hpp	Sparse tensors
Functional activations (riemann::functional::relu, gelu, softmax, …)	functions.hpp	Sparse tensors
Arithmetic operators (+ - * /)	engine/tensor/sparse_tensor.hpp	Arithmetic
save / load helpers	persistence.hpp	Training

Layers

All layers are LibTorch modules (TORCH_MODULE holders around ...Impl classes) in engine/nn/.

Family	Classes	Header
Convolution	RiemannConvolution, RiemannConvolutionTranspose, RiemannGenerativeConvolutionTranspose	engine/nn/conv.hpp
Normalization	RiemannBatchNorm, RiemannLayerNorm, RiemannGroupNorm, RiemannInstanceNorm, RiemannStableInstanceNorm	engine/nn/normalization.hpp
Activation	RiemannReLU, RiemannHardsigmoid, RiemannHardswish, RiemannSinusoidal	engine/nn/activation.hpp
Linear	RiemannLinear, RiemannDense	engine/nn/linear.hpp
Pooling	RiemannAvgPooling, RiemannSumPooling, RiemannMaxPooling, RiemannPoolingTranspose, RiemannGlobalPooling, RiemannGlobalAvgPooling, RiemannGlobalSumPooling, RiemannGlobalMaxPooling	engine/nn/pooling.hpp
Residual	BasicBlock, Bottleneck	engine/nn/residual.hpp
Squeeze-excite	SELayer, SEBasicBlock, SEBottleneck	engine/nn/squeeze_excite.hpp
Structure	RiemannUnion, RiemannPruning, RiemannToSparseTensor, RiemannToDenseTensor, RiemannToFeature, RiemannBroadcastAddition, RiemannBroadcastMultiplication, RiemannBroadcastConcatenation	engine/nn/operations.hpp
Interpolation	RiemannInterpolation	engine/nn/resample.hpp

Autograd functions behind the layers (for custom modules) are in autodiff/, for example autodiff/conv_func.hpp, autodiff/sparse_matmul_func.hpp (SPMMFunction) and autodiff/compat.hpp.

Training

Type	Header	Guide
training::OptimizerConfig, OptimizerType, StepLRScheduler	engine/learning/training.hpp	Training
training::SparseSample, training::SparseBatch	engine/learning/training.hpp	Training
StepResult, EpochResult, GradientAccumulator	engine/learning/training.hpp	Training

Adaptive voxelization

Type	Header	Guide
adaptive::AdaptiveSparseTensor, adaptive::SourcePoints	octree/tensor.hpp	Adaptive voxelization
adaptive::AdaptiveVoxelConfig, Reduction, RefinePolicy, AxisAlignedRegion	octree/config.hpp	Adaptive voxelization
adaptive::AdaptiveKernelConfig, AdaptiveConvMode, KernelMapCache	octree/stencil.hpp	Adaptive convolution
interpolate, InterpMode	octree/ops.hpp	Adaptive voxelization
voxelize_sphere, SphereFrame	octree/sphere.hpp	Adaptive voxelization

Gaussian splats

Type	Header	Guide
gaussian::GaussianModel	splat/model.hpp	Gaussian splats
gaussian::GaussianTensor	splat/soup.hpp	Gaussian splats
gaussian::Gaussian3D	splat/primitive.hpp	Gaussian splats
gaussian::Camera	splat/camera.hpp	Gaussian splats
gaussian::RenderOptions, RenderOutput, render	splat/render.hpp	Gaussian splats
gaussian::GaussianOptimizer	splat/optimizer.hpp	Gaussian splats
gaussian::GaussianTrainer, GaussianTrainerConfig	splat/trainer.hpp	Gaussian splats
PLY load/save	splat/ply.hpp	Gaussian splats
Spherical harmonics	splat/harmonics.hpp	Gaussian splats

Source

Browse the headers on GitHub. The reference is regenerated from them on every push to main (see Packaging).