Unpaired Translation of Point Clouds for Modeling Detector Response

TL;DR

The paper tackles detector-response modeling in time projection chambers by formulating it as unpaired translation between simulation-domain and experimental-domain point clouds. It integrates CycleDiffusion with a modified DPM-Encoder to translate events across domains without paired data, capturing both detector noise and response. Validations on synthetic point clouds and AT-TPC data show the method can learn conditional noise distributions and perform bidirectional translations with reliable reconstruction metrics, enabling more realistic simulators and improved noise handling in nuclear physics detectors. This approach offers a data-driven pathway to bridge simulations and experiments in complex 3-D detector environments.

Abstract

Modeling detector response is a key challenge in time projection chambers. We cast this problem as an unpaired point cloud translation task, between data collected from simulations and from experimental runs. Effective translation can assist with both noise rejection and the construction of high-fidelity simulators. Building on recent work in diffusion probabilistic models, we present a novel framework for performing this mapping. We demonstrate the success of our approach in both synthetic domains and in data sourced from the Active-Target Time Projection Chamber.

Figures (3)

• Figure 1: Schematic depicting our unpaired translation model at inference time. Arrows in red denote our modifications to the standard CycleDiffusion architecture to incorporate a point cloud diffusion model.
• Figure 2: Samples of translation results on on the ($\mathcal{L}_X, \mathcal{L}_Y$) dataset in two directions.
• Figure 3: Translation and reconstruction results on $(\mathcal{G}_X, \mathcal{G}_Y)$ (top) and $(\mathcal{A}_X, \mathcal{A}_Y)$ data (bottom).