Reproducibility materials for Full-Path Fixed-Channel Supervision for Compact Skin Lesion Segmentation
Year: 2026
Type: software
Abstract: Companion reproducibility materials for the PicoSemaNet/FPFCS manuscript. Includes model, training, evaluation and plotting code; the locked internal ISIC 2018 split identifiers (1815/259/520); original configurations, logs, GPU results, CPU Euclidean-BF1 re-evaluations, bootstrap inputs and result mappings; all 47 historical best checkpoints including the primary EC-free models and shared KD teacher; and best/last checkpoints for all six matched 100-epoch Full FPFCS versus No FPFCS runs at seeds 42, 43 and 44. The proposed configuration is segmentation + FPFCS + KD with EC disabled; EC remains an ablation. The original 632-file evidence unit is preserved unchanged. Original ISIC images and masks are not redistributed. New portable helpers and validation records are supplied; newly run CPU replays are identified separately from the original GPU experiment results. See README.md and REPRODUCE.md for precise protocol, metric, environment and historical-label details. This research was supported by Wenzhou-Kean University, China, through the IRSP project (Grant No. IRSPG202203). Licensing: original software is MIT; author-created experiment records, figures and weights are CC BY 4.0. See RIGHTS.md in the archive for category-specific scope; third-party terms remain unchanged. The MIT-licensed code repository is https://github.com/zykmin1726982627-cloud/PicoSemaNet (verified commit d16c386052b59ab86e28630b4ee1f99f093265f8; GitHub release v1.0.0). The repository provides a portable code release; this Zenodo archive preserves the original experimental evidence and the documented replay helpers. See the repository documentation for changes from the collected server sources.
Authors: Zheng, Yikuan, Fan, Bohe, Gupta, Gaurav
Institutions: Wenzhou-Kean University, Kean University
Language: English
Subfield: Oncology
Field: Medicine
Domain: Health Sciences
Open Access status: green
Funder: Wenzhou-Kean University
OpenAlex ID: W7223670004
