Note
This is a Hugging Face dataset. Learn how to load datasets from the Hub in the Hugging Face integration docs.
Dataset Card for neurips-2025-vision-papers#

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:
Analyze paper content using Vision-Language Models
Study trends in computer vision research
Build tools for academic paper understanding
Source Data#
Data Collection and Processing#
Initial Collection: Paper metadata scraped from NeurIPS 2025 virtual conference site
arXiv Matching: Papers matched with arXiv using title and author matching algorithms
Category Filtering: Filtered to include only vision-related categories (cs.CV, cs.MM, eess.IV, cs.GR, cs.RO) with valid arXiv IDs
PDF Download: First pages downloaded from arXiv (https://arxiv.org/pdf/{arxiv_id}.pdf)
Image Conversion: PDFs converted to PNG images at 500 DPI using pdf2image
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