Computational verification demonstrates model-specific statistical errors in join-based data pipelines across four super-resolution models, indicating the necessity of join-free validation.
Errata and supplement to Technical Report v1.10 and v1.12. Technical Report v1.12 corrected the Failure Fingerprint statistics (mu_stat, sigma_stat, and their deviation coordinates) for Upscayl Standard and SwinIR against an independent, join-free verification database, and explicitly deferred Upscayl High-Fidelity and Digital-Art to a later report. This report completes that remaining work for all four models studied in this series, re-deriving every value under the identical exact-methodology-matched procedure. Result: Upscayl Standard's sigma_stat is corrected from 0.0800 to 0.101 (confirming Technical Report v1.12); SwinIR is confirmed fully unaffected on both axes; Digital-Art is confirmed fully unaffected on both axes despite a 54.1% reduction in its underlying population count; Upscayl High-Fidelity is severely affected on both axes -- mu_stat corrected from 0.1400 to 0.0100, with the deviation coordinate collapsing from roughly 19x the theoretical constant to essentially on it, and sigma_stat corrected from 0.0800 to 0.0860. No two models show the identical impact pattern, confirming the bug's consequences are model-specific rather than a single fixable constant. As a secondary, independent cross-check, the SSIM=1.0 Pixel Identity transitivity corollary (every comp_* feature matching exactly between GT, noTTA, and TTA on blocks where both reconstructions score SSIM=1.0) is confirmed with zero exceptions across all four models, at population sizes ranging from 15 (High-Fidelity) to 2,657,100 (SwinIR). Along the way, a numerical coincidence between High-Fidelity's and SwinIR's corrected mu_stat values is investigated directly and resolved as a boundary-condition artifact of the split algorithm's own lower bound, not a genuine cross-model connection -- documented explicitly to prevent later mis-citation. Together, these results are framed as demonstrating two things at once: that the original join-based database construction carries real, model-specific risk rather than a single isolated defect, and that the new join-free pipeline is a validated correction, not merely a theoretical one -- alongside a reusable five-step procedure for trusting, or correctly distrusting, any aggregate statistic built on a join-based or estimator-based pipeline. Upscayl Standard's and SwinIR's databases are identical to those already independently verified and evidenced in Technical Report v1.12 and v1.12.1; this report cites that existing evidence directly and provides fresh hashes and screen-recorded video evidence only for Upscayl High-Fidelity and Digital-Art, new to this report. Dataset and Mandatory Citation:- Source: NIH ChestX-ray8 (Hospital-scale chest x-ray database)- Citation: Wang, X., Peng, Y., Lu, L., Lu, Z., Bagheri, M., & Summers, R. M. (2017). "ChestX-ray8: Hospital-scale chest x-ray database and benchmarks on weakly-supervised classification and localization of common thoracic diseases." Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 3462–3471. - Download: https://nihcc.app.box.com/v/ChestXray-NIHCC Contact : s.shiny.n.works@gmail.com
No takes yet. Share an insight, caveat, or question.
neco mohumohu (2026) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: