Week of 2026-06-08

pix2pix troubleshooting

Published

June 8, 2026

Overview

Mary reran the pix2pix model using a more tightly constrained dataset (following the preprocessing changes introduced after the initial run). Visual quality assessment is ongoing; a traceability gap in the output pipeline was identified and flagged for fixing.

Rerun Summary

  • Dataset: Constrained dataset (higher slice inclusion threshold, 512×512 with zero padding - see previous entry)
  • Checkpoints location: /home/lenkm/syn7T/pytorch-CycleGAN-and-pix2pix/checkpoints/ on LRI-111261
  • Output images: Located in the web/ subdirectory within each run folder

Visual Assessment

Difficult to determine from inspection alone how much the constrained dataset improved synthesis quality relative to the initial run. Formal quantitative evaluation (iBEAT v2.0 Dice/IoU) will be needed to resolve this.

Issues Identified

Slice traceability gap

During output review, it was noted that some paired 3T and 7T images did not appear to represent the same slice - suggesting possible misalignment or pairing errors in the dataset or preprocessing pipeline. The current pix2pix implementation does not save output images with a reference back to their source filenames, making it impossible to trace synthesized outputs back to the original input slices.

Planned fix: Modify the pix2pix code to embed or save source filename references alongside generated outputs. This will enable:

  • Quality assessment tied to specific input pairs
  • Identification of systematic misalignment patterns
  • Easier troubleshooting of artifact sources

Next Steps