SplatQ
A combinatorial optimization pruning algorithm for 3D Gaussian Splatting (3DGS) that jointly models deletion effects via spatial appearance blocking, budget quota smoothing, and Quadratic Unconstrained Binary Optimization (QUBO) formulation, enabling efficient selection of Gaussian primitives under a fixed storage or computation budget.
Timeline (last 90 days)
SplatQ is a combinatorial optimization pruning algorithm for 3D Gaussian Splatting (3DGS) that coordinates regional budget allocation and joint Gaussian selection under a fixed budget.
SplatQ partitions all Gaussians into size-constrained blocks based on spatial proximity and appearance similarity.
SplatQ smooths the retention quota for each block to prevent budgets from becoming overly concentrated in a small number of high-importance blocks.
SplatQ constructs a quadratic approximation model using the response of Gaussian opacity changes to representative pixel colors, incorporating three types of factors: the independent response of individual Gaussians, pairwise interactions between Gaussians, and quota penalty terms.
SplatQ integrates the pruning problem into a Quadratic Unconstrained Binary Optimization (QUBO) objective function, where the retention or deletion of each Gaussian corresponds to a binary variable.
SplatQ solves the QUBO objective using two approaches: simulated annealing (SA) and coherent Ising machines (CIM).
The research team tested on 5 benchmark datasets, and SplatQ outperformed current state-of-the-art pruning methods across three dimensions: aggregate reconstruction quality, spatial quality balance, and rendering speed.
Ablation experiments confirm that budget smoothing improves local detail reconstruction quality, and interaction modeling improves the overall quality of Gaussian selection under a fixed budget.
All Facts (8)
SplatQ is a combinatorial optimization pruning algorithm for 3D Gaussian Splatting (3DGS) that coordinates regional budget allocation and joint Gaussian selection under a fixed budget.
50%UnverifiedSplatQ partitions all Gaussians into size-constrained blocks based on spatial proximity and appearance similarity.
50%UnverifiedSplatQ smooths the retention quota for each block to prevent budgets from becoming overly concentrated in a small number of high-importance blocks.
50%UnverifiedSplatQ constructs a quadratic approximation model using the response of Gaussian opacity changes to representative pixel colors, incorporating three types of factors: the independent response of individual Gaussians, pairwise interactions between Gaussians, and quota penalty terms.
50%UnverifiedSplatQ integrates the pruning problem into a Quadratic Unconstrained Binary Optimization (QUBO) objective function, where the retention or deletion of each Gaussian corresponds to a binary variable.
50%UnverifiedSplatQ solves the QUBO objective using two approaches: simulated annealing (SA) and coherent Ising machines (CIM).
50%UnverifiedThe research team tested on 5 benchmark datasets, and SplatQ outperformed current state-of-the-art pruning methods across three dimensions: aggregate reconstruction quality, spatial quality balance, and rendering speed.
50%UnverifiedAblation experiments confirm that budget smoothing improves local detail reconstruction quality, and interaction modeling improves the overall quality of Gaussian selection under a fixed budget.
50%