Saved in:
Bibliographic Details
Main Authors: Barnhill, Alexander, Towers, Jared
Format: Recurso digital
Language:
Published: Zenodo 2026
Online Access:https://doi.org/10.5281/zenodo.18363012
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866901668923179008
author Barnhill, Alexander
Towers, Jared
author_facet Barnhill, Alexander
Towers, Jared
contents <h1>FinFriend</h1> <p><strong>Encounter-conditioned fusion for photo-identification</strong> (taxon-agnostic; case study: Bigg’s killer whales)</p> <p>FinFriend is an <strong>encounter-aware photo-identification fusion pipeline</strong> that refines per-image identity posteriors using two lightweight context terms learned from the training split: (i) <em>global sighting priors</em> and (ii) an <em>encounter-conditioned co-occurrence (log-lift) context prior</em>. The method is model-agnostic and operates as post-processing on classifier outputs.</p> <h2>Zenodo contents (review/repro bundle)</h2> <ul> <li><code>code/</code> — source snapshot used in the manuscript</li> <li><code>splits/</code> — precomputed chronological splits + TRAIN-only artifacts (priors, co-occurrence, loglift)</li> <li><code>pkl/</code> — per-image classifier outputs (class map + logits/probabilities)</li> </ul> <h2>Reproduce manuscript results</h2> <ol> <li>Set <code>root</code> in <code>configs/paths.yaml</code> to the directory containing <code>code/</code>, <code>splits/</code>, and <code>pkl/</code>.</li> <li>Install dependencies: <code>pip install -r requirements.txt</code></li> <li>Run: <code>python -m inference.assisted_predictor</code></li> </ol> <p><strong>No leakage invariant:</strong> co-occurrence artifacts (priors/loglift) are computed from TRAIN only, and the test split is the newest encounters (chronological split).</p> <p>For full documentation, configuration details (Hydra), and optional training code, see the included <code>README</code>.</p>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_18363012
institution Zenodo
language
publishDate 2026
publisher Zenodo
record_format zenodo
spellingShingle Encounter-conditioned photo-ID fusion (FinFriend code + data release)
Barnhill, Alexander
Towers, Jared
<h1>FinFriend</h1> <p><strong>Encounter-conditioned fusion for photo-identification</strong> (taxon-agnostic; case study: Bigg’s killer whales)</p> <p>FinFriend is an <strong>encounter-aware photo-identification fusion pipeline</strong> that refines per-image identity posteriors using two lightweight context terms learned from the training split: (i) <em>global sighting priors</em> and (ii) an <em>encounter-conditioned co-occurrence (log-lift) context prior</em>. The method is model-agnostic and operates as post-processing on classifier outputs.</p> <h2>Zenodo contents (review/repro bundle)</h2> <ul> <li><code>code/</code> — source snapshot used in the manuscript</li> <li><code>splits/</code> — precomputed chronological splits + TRAIN-only artifacts (priors, co-occurrence, loglift)</li> <li><code>pkl/</code> — per-image classifier outputs (class map + logits/probabilities)</li> </ul> <h2>Reproduce manuscript results</h2> <ol> <li>Set <code>root</code> in <code>configs/paths.yaml</code> to the directory containing <code>code/</code>, <code>splits/</code>, and <code>pkl/</code>.</li> <li>Install dependencies: <code>pip install -r requirements.txt</code></li> <li>Run: <code>python -m inference.assisted_predictor</code></li> </ol> <p><strong>No leakage invariant:</strong> co-occurrence artifacts (priors/loglift) are computed from TRAIN only, and the test split is the newest encounters (chronological split).</p> <p>For full documentation, configuration details (Hydra), and optional training code, see the included <code>README</code>.</p>
title Encounter-conditioned photo-ID fusion (FinFriend code + data release)
url https://doi.org/10.5281/zenodo.18363012