Uloženo v:
| Hlavní autor: | |
|---|---|
| Médium: | Recurso digital |
| Jazyk: | |
| Vydáno: |
Zenodo
2026
|
| Témata: | |
| On-line přístup: | https://doi.org/10.5281/zenodo.19701134 |
| Tagy: |
Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
|
Obsah:
- <p>First release of the repository. Contains two notebooks that together constitute a <strong>FAIR computational reproduction</strong> of <a href="https://doi.org/10.3389/fmars.2025.1699781">Decrop et al. 2025</a> plus a downstream artefact used by <a href="https://github.com/annefou/fiesta-scattering-bio">fiesta-scattering-bio</a>.</p> <h2>Notebooks</h2> <p>| Notebook | Input | Output | Purpose | |---|---|---|---| | `01_reproduce_decrop.py` | `test.txt` (33,718 images) | `results/reproduce_decrop_results.json` + `reproduce_decrop_predictions.npz` | Reproduce the five published metrics | | `02_cnn_val_predictions.py` | `val.txt` (33,829 images) | `results/cnn_predictions_val.npz` | Provide CNN predictions on the held-out val split for downstream stacking in fiesta-scattering-bio |</p> <p>Both notebooks share an <strong>identical pipeline</strong> — same pretrained EfficientNetV2-B0 weights, same preprocessing, same 10-crop test-time augmentation. Only the split differs.</p> <h2>Reproduction outcome on `test.txt`</h2> <p>Every reported metric matches the paper to within <strong>0.003 percentage points</strong>:</p> <p>| Metric | Paper | This reproduction | Delta (pp) | |---|---:|---:|---:| | Top-1 accuracy | 86.34 % | <strong>86.3426 %</strong> | +0.003 | | Top-5 accuracy | 98.76 % | <strong>98.7633 %</strong> | +0.003 | | Micro F1 | 86.34 % | <strong>86.3426 %</strong> | +0.003 | | Macro F1 | 78.76 % | <strong>78.7583 %</strong> | −0.002 | | Weighted F1 | 86.25 % | <strong>86.2470 %</strong> | −0.003 |</p> <p>Inference took <strong>27 minutes</strong> on a single M1 Pro CPU pool. All inputs are openly archived: dataset (Zenodo 10554845), weights (Zenodo 15269453), code (planktonclas, Apache 2.0). A third party can rerun this end-to-end with zero guessing.</p> <h2>What this release enables</h2> <ul> <li><strong>Versioned Zenodo DOI</strong> for this reproduction, citable alongside Decrop et al. 2025.</li> <li><strong>Docker image</strong> on GHCR: `ghcr.io/annefou/fiesta-decrop-reproduction:0.2.0` — runs the whole pipeline in one `docker run`.</li> <li><strong>CNN prediction artefacts</strong> (`cnn_predictions_test.npz`, `cnn_predictions_val.npz`) consumable by the stacked CNN + scattering experiment in <a href="https://github.com/annefou/fiesta-scattering-bio">fiesta-scattering-bio</a> v0.3.0.</li> </ul> <h2>FIESTA-OSCARS context</h2> <p>Part of the five-repo <a href="https://oscars-project.eu/projects/fair-image-analysis-across-sciences">FIESTA-OSCARS</a> project. This repo plays two roles:</p> <ol> <li>Standalone FAIR reproduction of a published CNN classifier.</li> <li>Upstream dependency for the biodiversity scattering-stacking experiment.</li> </ol> <p> Generated with <a href="https://claude.com/claude-code">Claude Code</a></p>