Recoverable Identifier

arXiv:2605.09203 · detector doi_compliance · incontrovertible · 2026-05-19 10:28:33.170083+00:00

Paper page Integrity report arXiv Try DOI

Evidence text

Jiren Zhu, Russell Kaplan, Justin Johnson, and Li Fei-Fei. HiDDeN: Hiding data with deep networks. InProceedings of the European Conference on Computer Vision (ECCV), pages 682–697, Cham, Switzerland, 2018. Springer. doi: 10.1007/978-3- 030-01267-0_40. A Generative AI Usage We used Claude (Anthropic) and ChatGPT (OpenAI) during the preparation of this paper. These tools assisted with phrasing and grammar edits in the main text, and with drafting portions of the training, evaluation, and plotting code. All AI-suggested text and code were reviewed by the authors before inclusion, and the re- ported experimental results were produced by scripts the authors ran end-to-end. The authors take full responsibility for the correct- ness of all claims, results, and references in this paper. B Training and Evaluation Details Section 3.2 lists the core training hyperparameters. This appendix fills in the remaining configuration details a reader would need to reproduce the numbers in Table 3 exactly. All six detectors share the same configuration; only the dataset differs. The optimizer is AdamW with weight decay10 −4, label smoothing is0 .1, and the learning rate follows the linear warmup already described and then decays to zero on a cosine schedule over the remaining epochs. Training runs for up to50epochs; early stopping usually triggers between epochs10and25depending on dataset size. We do not set a global random seed, so results are not bitwise reproducible across runs. Training used

Evidence payload

{
  "printed_excerpt": "Jiren Zhu, Russell Kaplan, Justin Johnson, and Li Fei-Fei. HiDDeN: Hiding data with deep networks. InProceedings of the European Conference on Computer Vision (ECCV), pages 682\u2013697, Cham, Switzerland, 2018. Springer. doi: 10.1007/978-3- 030",
  "reconstructed_doi": "10.1007/978-3-030-01267-0_40.A",
  "ref_index": 51,
  "resolved_title": null,
  "verdict_class": "incontrovertible"
}