Dataset class for GSC with scene-specific train/test view split.

A simple dataset class.

COLMAP parser.

Field-specific quantization parameters.

Quantization controls for a single attribute.

Camera pose optimization module.

A default strategy that follows the original 3DGS paper: `3D Gaussian Splatting for Real-Time Radiance Field Rendering <https://arxiv.org/abs/230

Configuration for adaptive mask stage.

Uses quantization and sorting to compress splats into PNG files and uses K-means clustering to compress the spherical harmonic coefficents. .

Record describing a single pruning step.

Statistics required to dequantize a single tensor field.

Strategy that follows the paper: `3D Gaussian Splatting as Markov Chain Monte Carlo <https://arxiv.org/abs/2404.09591>`_ This strategy will:

Permutation metadata produced by the mapping stage.

Metadata storing quantization statistics for every field.

Appearance optimization module.

Lower bound operator, computes `torch.max(x, bound)` with a custom gradient. The derivative is replaced by the identity function when `x` is

No-op simulation used when compression simulation is disabled.

COLMAP parser.

Class for 3D bilateral grids. Holds one or more than one bilateral grids.

Top-level configuration for compression simulation.

Uses quantization and sorting to compress splats into PNG files and uses K-means clustering to compress the spherical harmonic coefficents. .

Configuration for entropy constraint stage.

Uses quantization and sorting to compress splats into mp4 files via libx265 and uses K-means clustering to compress the spherical harmonic coeff

Specification describing how to locate post-training splats.

Adapter that wraps the legacy CompressionSimulation implementation.

A modified densification and pruning strategy of spacetime gaussian paper:

Metadata describing the result of the pruning stage.

Configuration for quantization stage.

Configuration for differentiable quantization.

Uses quantization and sorting to compress splats into PNG files and uses K-means clustering to compress the spherical harmonic coefficents. .

A Densification and Pruning strategy that follows the spacetime gaussian paper:

Uses quantization and sorting to compress splats into mp4 files via libx265 and uses K-means clustering to compress the spherical harmonic coeff

Uses quantization and sorting to compress splats into mp4 files via libx265 and uses K-means clustering to compress the spherical harmonic coeff

A generic command-line wrapper for various video codecs. This class provides a unified interface for different codecs (e.g., vvenc, VTM,

Handles saving tensor data to YUV files with metadata, and loading it back. This class encapsulates the logic for different attribute types,

Compression simulation built on top of the refactored runtime components.

INVR parser.

Single-frame orchestrator for post-training compression.

Configuration for preprocessing transforms (log transform, quat normalization, etc.).