About this resource
<p>Reproducibility package for "Corruption Robustness as a Distribution Shift Proxy: Cross-Dataset Predictive Validity".</p> <p>This release contains the reference code and artifacts used to reproduce the reported experiments, including:</p> <ul> <li>ImageNet-C <strong>BRI</strong> / mCE computation and <strong>ABTP</strong> on ImageNet-C and Oxford-IIIT Pets (<code>scripts/</code> with numbered phase drivers; shared code under <code>src/</code>),</li> <li><strong>15-model</strong> generalization evaluation, sensitivity / LOMO analysis, and figure generation from JSON,</li> <li><strong>Pre-computed summary JSON</strong> under <code>results/</code> (where shipped in this repository),</li> <li><strong>Quality checks</strong> (<code>qc_phase1a.py</code>, <code>qc_phase3.py</code>) and a <strong>Spearman verification snippet</strong> in <code>REPRODUCE.md</code>.</li> </ul> <p>Standard vision benchmarks and Oxford-IIIT Pets are <strong>not</strong> redistributed; users must obtain data under the respective licenses and set the documented environment variables.</p> <p>All reported tables and figures can be reproduced by following <strong><code>REPRODUCE.md</code></strong> (environment setup, script order, and optional correlation check). There is no single entrypoint script; phases are run as documented.</p> <p>This release is intended as the archival version associated with the paper submission.</p>
Files and access
Resources are read online through DARE. Direct file links are not published; access is checked against DSpace when a document is opened. Authorised librarian workflows are unaffected.
This record contains bibliographic metadata, but no files are currently attached.
