Note

This is a Hugging Face dataset. Learn how to load datasets from the Hub in the Hugging Face integration docs.

Hugging Face

Dataset Card for neurips-2025-vision-papers#

image/png

This is a FiftyOne dataset with 1134 samples.

Installation#

If you haven’t already, install FiftyOne:

pip install -U fiftyone

Usage#

import fiftyone as fo
from fiftyone.utils.huggingface import load_from_hub

# Load the dataset
# Note: other available arguments include 'max_samples', etc
dataset = load_from_hub("Voxel51/visual_ai_at_neurips2025")

# Launch the App
session = fo.launch_app(dataset)

Dataset Details#

Dataset Description#

This dataset contains NeurIPS 2025 accepted papers focused on computer vision and related fields, enriched with arXiv metadata and first-page images. It includes papers from multiple vision-related categories including Computer Vision (cs.CV), Multimedia (cs.MM), Image and Video Processing (eess.IV), Graphics (cs.GR), and Robotics (cs.RO). Each entry includes paper metadata, abstracts, author information, and a high-resolution (500 DPI) PNG image of the paper’s first page.

  • Curated by: Harpreet Sahota

  • Language(s) (NLP): en

  • License: Apache 2.0

Dataset Sources#

  • Original Data Source: NeurIPS 2025 Conference (https://neurips.cc/virtual/2025/calendar)

  • arXiv API: https://arxiv.org/

Uses#

Direct Use#

This dataset is suitable for:

  • Analyzing trends in computer vision research at NeurIPS 2025

  • Vision-Language Model (VLM) analysis of paper content

  • OCR and text extraction from academic papers

  • Building search and recommendation systems for academic papers

  • Studying paper formatting, structure, and visual presentation

  • Training models to understand academic paper layouts

Out-of-Scope Use#

This dataset should not be used for:

  • Representing the complete NeurIPS 2025 corpus (only vision-related papers with arXiv IDs)

  • Papers without arXiv IDs are not included

  • Full paper content analysis (only first pages are included)

  • Citation analysis (references are not included)

Dataset Structure#

The dataset contains the following fields:

  • filepath: Path to the first-page PNG image (500 DPI)

  • type: Paper presentation type (e.g., “Poster”, “Oral”)

  • name: Paper title

  • virtualsite_url: URL to the paper on NeurIPS virtual site

  • abstract: Paper abstract

  • arxiv_id: arXiv identifier (e.g., “2301.12345v2”)

  • arxiv_authors: List of paper authors from arXiv

  • arxiv_category: Classification field with paper category (cs.CV, cs.MM, eess.IV, cs.GR, or cs.RO)

Dataset Creation#

Curation Rationale#

This dataset was created to provide a focused collection of vision-related papers from NeurIPS 2025 with high-quality first-page images for multimodal analysis. The motivation was to enable researchers and practitioners to:

  1. Analyze paper content using Vision-Language Models

  2. Study trends in computer vision research

  3. Build tools for academic paper understanding

Source Data#

Data Collection and Processing#

  1. Initial Collection: Paper metadata scraped from NeurIPS 2025 virtual conference site

  2. arXiv Matching: Papers matched with arXiv using title and author matching algorithms

  3. Category Filtering: Filtered to include only vision-related categories (cs.CV, cs.MM, eess.IV, cs.GR, cs.RO) with valid arXiv IDs

  4. PDF Download: First pages downloaded from arXiv (https://arxiv.org/pdf/{arxiv_id}.pdf)

  5. Image Conversion: PDFs converted to PNG images at 500 DPI using pdf2image

  6. Quality: 500 DPI ensures readability of 10pt font common in academic papers

Who are the source data producers?#

  • NeurIPS 2025 Conference: Original paper metadata and acceptance decisions

  • arXiv: Paper PDFs and metadata

  • Paper Authors: Original paper content

Annotations#

Annotation process#

The arxiv_category field represents the primary arXiv category assigned by paper authors during submission. No additional manual annotations were added.

Bias, Risks, and Limitations#

Limitations:

  • Only includes papers with arXiv IDs (some NeurIPS papers may not be on arXiv)

  • Only includes first page (no full paper content)

  • Limited to specific vision-related categories

  • arXiv matching may have errors or mismatches

  • Images are high resolution (500 DPI) resulting in larger file sizes

Biases:

  • Excludes papers without arXiv presence

  • May underrepresent certain research areas or institutions with different publication practices

  • Category classification reflects author self-assignment on arXiv

Recommendations#

Users should be made aware that:

  • This is not a complete representation of NeurIPS 2025 papers

  • arXiv matching was automated and may contain errors

  • Only first pages are available (for full papers, refer to arXiv or NeurIPS proceedings)

  • High DPI images require significant storage space

Citation#

NeurIPS 2025:

@inproceedings{neurips2025,
  title={Neural Information Processing Systems},
  year={2025},
  organization={NeurIPS}
}

Dataset Card Contact#

Harpreet Sahota