version: "1.0.0" name: update-pcc-papers description: > Fetch the latest learned point cloud compression (PCC) papers from arXiv, IEEE, CVF Open Access and other sources, deduplicate them against the existing README, and append new entries to the correct year section. Use when the user asks to "update papers", "pull latest literature", "find new PCC papers", "refresh the awesome list", or any time the awesome-learned-point-cloud-compression repo needs its literature list brought up to date.

Update Learned Point Cloud Compression Papers

This skill keeps the awesome-learned-point-cloud-compression README up to date with the newest learned / neural point cloud compression literature.

Target file

README.md at the repository root.

Entry format (MUST follow exactly)

Each paper is a single bullet line under a ### <year> heading inside the ## Papers section:

-[[Venue](link)] Paper Title. [[Code](repo-url)]

Rules:

• Venue = short venue tag (see reference.md for the full list).
• If a preprint has no venue yet, use [[arxiv](arxiv-url)].
• [[Code](repo-url)] is optional — only include when an official / author

GitHub repo exists.

• One blank line between consecutive entries.
• Entries inside a year are ordered newest-first (most recent publication date on top).

Workflow

Step 1 — Read the current state

1. Read README.md.
2. Collect every existing paper title (lower-cased) into a set for duplicate

checking.

Step 2 — Search for new papers

Run parallel web searches using the query templates below. Use the current year and the previous year. Add more years if the user requests a wider window.

Search query templates (replace {YEAR}):

Additionally, fetch the arXiv listing pages for cs.CV / eess.IV filtered by the keyword point cloud compression: https://arxiv.org/list/cs.CV/{YEAR}{MM} for recent months (only the last 6 months are needed for an update).

Step 3 — Filter and deduplicate

For each candidate paper found:

1. Relevance check — the paper must be primarily about *learned / neural /

deep-learning-based* point cloud compression (geometry, attribute, or joint).

• Exclude pure MPEG G-PCC / V-PCC traditional codec papers unless they

contain a learned component.

• Exclude point cloud quality assessment or enhancement papers that do

not propose a compression method (unless they are already represented in the repo — keep parity with existing style).

1. Duplicate check — compare the lower-cased title against the existing set

(Step 1). Skip if already present. Use fuzzy match: if >85 % of significant words overlap, treat as duplicate.

1. Venue resolution — try to identify the official venue:

• Check CVF Open Access, IEEE Xplore, ACM DL, openreview, the paper PDF.
• If only an arXiv version exists, tag as arxiv.

1. Code link — search for an author GitHub repo (check the paper PDF

"Code / Project page" footnote, and search GitHub by paper title). Only add [[Code](...)] when a repo is found.

Step 4 — Format entries

Build each new entry string following the format in the "Entry format" section above.

Step 5 — Insert into README

1. Determine the publication year of each new paper.
2. Find or create the corresponding ### {year} subsection under ## Papers.

• If the year section does not exist yet, create it immediately above

the next-most-recent year section (years are sorted descending).

1. Insert new entries at the top of that year section (newest first).
2. Preserve all existing entries and formatting.

Step 6 — Summarise for the user

After editing, print a short summary table:

| Year | Venue | Title | Code |
|------|-------|-------|------|

Also list any candidates that were skipped and the reason (duplicate / not learned-based / venue not found).

Key venue short-tags

Common venues already used in this repo:

CVPR, ICCV, ECCV, NeurIPS/NIPS, AAAI, ICML, IJCAI, ACM MM, ACM TOMM, TPAMI, TIP, TMM, TCSVT, TVCG, TII, RA-L, ICASSP, ICME, ICIP, DCC, VCIP, PCS, MMVE, MM Asia, ICRA, IROS, ICMR, APCCPA, CMM, CAS, EG, IET, arxiv

For the full venue list and search-URL cheat-sheet, see reference.md.

Important rules

• Never invent links. Every URL must be verified by actually fetching the

page or seeing it in search results.

• Never fabricate paper titles. Only add papers that were found in search

results or arXiv listings.

• Keep alphabetical / chronological parity with the existing style.
• When unsure whether a paper qualifies as "learned", default to excluding it

and mention it in the skipped summary so the user can decide.

• After editing, do not run git commit unless the user explicitly asks.