Transformers Improve Subpixel Position Reconstruction in Segmented Scintillator
A study posted Dec. 12, 2025, demonstrates that transformer-based neural networks outperform centroid-based methods in reconstructing subpixel positions in an 8×8 segmented scintillator detector using Geant4-simulated cosmic ray muons. The transformer achieved 1.14° RMS angular error and 0.24 cm position MAE—improvements of 2.22× and 6.33× over energy-weighted centroids—enabling more precise muon tomography and cosmic-ray trajectory reconstruction.
Key Points
- 1Report transformers achieve 1.14° RMS angular error and 0.24 cm position MAE
- 2Demonstrate 2.22× angular and 6.33× positional improvement over energy-weighted centroid baseline
- 3Enable higher-resolution muon tomography and cosmic-ray tracking for detector calibration and event reconstruction
Scoring Rationale
Strong empirical improvements and practical reconstruction method; limited by simulation-only results, single-detector geometry, and preprint status.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
