Tracing the Oracle: Improving Diffusion Timestep Scheduling for 3D CT Reconstruction
For practitioners of 3D CT reconstruction, this work addresses the computational overhead and truncation errors of diffusion models by providing an optimized timestep schedule, though it is an incremental improvement over existing heuristic schedules.
The paper proposes Tracing the Oracle (TrO), a plug-and-play framework that optimizes diffusion timestep scheduling for 3D CT reconstruction by minimizing cumulative error against a reference oracle, achieving improved reconstruction fidelity and computational efficiency under a strict budget of ≤10 sampling steps.
Pretrained diffusion models demonstrate impressive potential in solving highly ill-posed 3D computed tomography (CT) inverse problems, while the inference process suffers from significant computational overhead. Furthermore, existing uniform timestep schedules fail to capture the non-uniform evolution of the reverse conditional diffusion stochastic differential equation, thereby introducing substantial truncation errors. To overcome this limitation, we propose Tracing the Oracle (TrO), a plug-and-play framework for improved timestep scheduling. Specifically, we treat densely sampled numerical integration trajectories on a few samples as the reference oracle. The optimized schedule is extracted by leveraging dynamic programming to globally minimize the cumulative error between the few-step approximation and the oracle. This mechanism precisely allocates the limited sampling steps to critical evolution stages that are highly susceptible to truncation errors. Our extensive experiments on the AAPM dataset across multiple 3D CT reconstruction tasks demonstrate that, when combined with the state-of-the-art 3D CT reconstruction method DDS, our optimized timesteps significantly improve reconstruction fidelity and computational efficiency compared to existing heuristic schedules, especially under a strict budget of no more than 10 sampling steps.