Datasets:

---

LG-AI-Research

/

PANORAMA

like 3

Follow

LG AI Research 17

This is an unofficial, research-oriented extension and not affiliated with the
original authors

Hi PANORAMA authors and community — first, thank you for this dataset. The
decision-trail framing and the claim-level NOC4PC task are genuinely useful.

While working with NOC4PC I noticed the task is text-only, even though real
examiners rely heavily on the drawings of both the application and the cited
prior art. So I built a multimodal extension and released it (with full credit to
your work, under the same CC-BY-NC-4.0 license):

👉 https://huggingface.co/datasets/sungjae98/PANORAMA-NOC4PC-multimodal

What it adds

• The drawings for both the patent application and*every cited prior-art
  reference, rendered from the original USPTO PDFs (120 dpi, all pages).
• Each NOC4PC row gets two new columns — application_image_ids and
  prior_art_image_ids — so a model can see the figures the examiner compared.

Structure (reference layout, like VQAv2 / DocVQA)
Because the same drawing is shared by many claim-level rows, embedding per row
would balloon the dataset past 600 GB. Instead I store each unique drawing once
and reference it by image_id:

• config noc4pc — original fields + application_image_ids[] + prior_art_image_ids[]
  (train 136,211 / val 7,392 / test 2,884)
• config drawings — 298,935 unique drawing pages with datasets.Image (≈ 39 GB)