Seeking arXiv cs.CV endorsement. Detector that recovers interactive form fields in flat PDFs, weights are on the Hub

Independent researcher Samuel A. in Germany is seeking arXiv cs.CV endorsement for AcroMELD, a 39.4M parameter detector that recovers interactive form fields in flat PDFs, with weights available on Hugging Face Hub and inference package acromeld on PyPI. The paper reports failure modes including a complete transfer failure on the signature class, and was evaluated once against a sealed external holdout under a pre-registered, hash-bound protocol.

Hi everyone, I’m an independent researcher in Germany working on PDF accessibility. I’ve just finished a paper on AcroMELD, a 39.4M parameter detector that recovers the interactive form fields in PDFs that look like forms but aren’t actually fillable. It combines a compact visual transformer with label-free PDF drawing structure as a second modality, and it was evaluated exactly once against a sealed external holdout under a pre-registered, hash-bound protocol. The paper reports the failure modes just as openly as the wins, including a complete transfer failure on the signature class. Everything around the paper is public and checkable. The trained weights live right here on the Hub: Cadmon/AcroMELD · Hugging Face https://huggingface.co/Cadmon/AcroMELD and the matching inference package is called acromeld on PyPI. I was also a co-author on a cs.CV arXiv paper back in 2019 arXiv:1906.09587, “Semi-Supervised Learning for Cancer Detection of Lymph Node Metastases” , but that submission ran through a colleague’s account and is older than the five year window arXiv counts for endorsement qualification. So my own account still needs a cs.CV endorsement before I can submit. If you’re qualified to endorse for cs.CV arXiv shows this on any of your papers via the link “Which of the authors of this article can endorse?” , please send me a DM or reply here and I’ll share the PDF and my endorsement code with you privately. Endorsing only confirms that the topic fits the category. It is not a review and doesn’t tie your name to the paper. Happy to answer anything about the paper in this thread. Thanks for reading. Samuel A.