Populate Paper Metadata

Check the meta_missing and summary_issue fields to see exactly what needs to be filled for each paper.

Step 1: Gather Information

To accurately categorize the paper, gather its context:

1. Read the text: Read the paper's ## LLM Summary and abstract: from the markdown file.
2. Fetch source: For higher accuracy, you should fetch the arXiv abstract page via https://arxiv.org/html/{arxiv_id} or https://arxiv.org/abs/{arxiv_id}.
3. Review existing tags: Look at the INDEX.md or tags/ folder to see what standard tags already exist. Try to reuse existing standard tags before inventing new ones (e.g., use posed-multi-view-images instead of posed-images).
4. Identify papers: Check the INDEX.md or list the papers/ directory to identify the filenames of related or compared papers.

Step 2: Determine Rules for Each Field

Human-only fields — do NOT fill: inputs, outputs, methods. These must only be filled by a human who has read the paper. Leave them as empty placeholders (- ).

benchmarks ✅ auto-fillable

• Definition: Which datasets are used to evaluate the method?
• Format: List of kebab-case tags.
• Examples: mipnerf360, tanks-and-temples, deep-blending, waymo, nerf-synthetic.

related ✅ auto-fillable

• Definition: Which existing papers in our knowledge base does this paper build upon, extend, or take direct inspiration from?
• Format: List of exact filenames without the .md extension.
• Rule: You MUST search the papers/ directory. Only include a related paper if it already exists in our knowledge base.
• Example: 2308_3d-gaussian-splatting

compared ✅ auto-fillable

• Definition: Which prior works does this paper compare its results against?
• Format: List of exact filenames without the .md extension.
• Rule: You MUST search the papers/ directory. Only include a paper if it already exists in our knowledge base.
• Example: 2003_nerf, 2201_instant-ngp

Step 3: Write the LLM Summary

If the paper's summary_issue from Step 0 is not null (e.g. empty or no-structure), you must write or re-write the ## LLM Summary section in the markdown body.

Guidelines:

• Write in third person, technical tone.
• Open with a one-sentence summary of the core contribution.
• List key contributions as bullet points with bold labels.
• Include quantitative results if notable (e.g., speed, PSNR).
• Target length: 100-200 words.
• Must end with: > *Auto-generated summary. Do not edit manually.*

Step 4: Populate the File

Edit the paper's markdown file to replace the empty arrays with your findings and insert the ## LLM Summary.

Before: