README.md

---
language:
- en
license: mit
task_categories:
- text-classification
- question-answering
- summarization
- text-retrieval
tags:
- computer-vision
- paper
- cvpr
- ocr
- pdf
- research
size_categories:
- 10K<n<100K
---

# CVPR Papers

<p align="center">
  <img src="images/cvpr.jpeg" alt="CVPR Papers" width="800"/>
</p>

<p align="center">
  <a href="https://github.com/choucisan/CVpaper"><img src="https://img.shields.io/badge/GitHub-CVPR_Papers-181717?style=for-the-badge&logo=github" alt="GitHub"></a>
  <a href="https://huggingface.co/datasets/choucsan/CVPR_Papers"><img src="https://img.shields.io/badge/%F0%9F%A4%97_HuggingFace-Dataset-yellow?style=for-the-badge" alt="Hugging Face"></a>
  <a href="https://choucisan.github.io/collections/cvpr_paper"><img src="https://img.shields.io/badge/Blog-Post-blue?style=for-the-badge" alt="Blog"></a>
  <a href="https://choosealicense.com/licenses/mit"><img src="https://img.shields.io/badge/License-MIT-green?style=for-the-badge" alt="License"></a>
</p>

Since 2013, deep learning has revolutionized computer vision, with CVPR (IEEE Conference on Computer Vision and Pattern Recognition) serving as the premier venue documenting this transformative journey. From AlexNet's breakthrough to the rise of Transformers, CVPR papers chronicle the complete trajectory of computer vision advancement.

**CVPR Papers** is a comprehensive dataset containing all papers from CVPR 2013 to present, including metadata and PDF files. It is designed for literature review, trend analysis, citation network construction, and various research tasks in computer vision.

---

## Pipeline

The dataset is constructed through a systematic multi-stage pipeline:

1. **Web Scraping**: Extract paper listings from CVF Open Access repository
2. **Metadata Extraction**: Parse HTML to extract titles, authors, PDF links, and BibTeX citations
3. **Abstract Retrieval**: Fetch abstracts from individual paper detail pages
4. **PDF Download**: Concurrently download all paper PDF files
5. **Data Validation**: Verify data integrity and format consistency

---

## Dataset Structure

```
CVPR_Papers/
├── 2013/
│   ├── pdf/                    # PDF files for all papers
│   │   ├── paper1.pdf
│   │   ├── paper2.pdf
│   │   └── ...
│   └── meta.jsonl         # Metadata 
│ 
├── 2014/
│   ├── pdf/
│   └── meta.jsonl
└── ...
```

---

## Dataset Overview

- **Total Papers**: 22,521 (CVPR 2013-2026, continuously expanding)
- **Data Format**: JSONL for metadata, PDF for full papers, organized by year
- **Source**: [CVF Open Access](https://openaccess.thecvf.com)

| Field | Type | Description |
|-------|------|-------------|
| `title` | string | Paper title |
| `authors` | string | Comma-separated list of authors |
| `abstract` | string | Paper abstract |
| `pdf_path` | string | Relative path to PDF file |
| `download_url` | string | Direct download URL for PDF |
| `bibtex` | string | BibTeX citation string |

**Example Entry:**
```json
{
  "title": "Deformable Spatial Pyramid Matching for Fast Dense Correspondences",
  "authors": "Jaechul Kim, Ce Liu, Fei Sha, Kristen Grauman",
  "abstract": "We introduce a fast deformable spatial pyramid (DSP) matching algorithm for computing dense pixel correspondences...",
  "pdf_path": "2013/pdf/Kim_Deformable_Spatial_Pyramid_2013_CVPR_paper.pdf",
  "download_url": "https://openaccess.thecvf.com/content_cvpr_2013/papers/Kim_Deformable_Spatial_Pyramid_2013_CVPR_paper.pdf",
  "bibtex": "@InProceedings{Kim_2013_CVPR,author = {Kim, Jaechul and Liu, Ce and Sha, Fei and Grauman, Kristen},title = {Deformable Spatial Pyramid Matching for Fast Dense Correspondences},booktitle = {Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},month = {June},year = {2013}}"
}
```

---

## Quick Start

### Installation

```bash
pip install huggingface_hub requests
```

### Load Metadata

```python
from huggingface_hub import hf_hub_download
import json

# Download metadata for a specific year
year = "2013"
meta_path = hf_hub_download(
    repo_id="choucsan/CVPR_Papers",
    filename=f"{year}/meta.jsonl",
    repo_type="dataset",
)

# Load metadata
papers = []
with open(meta_path, 'r', encoding='utf-8') as f:
    for line in f:
        papers.append(json.loads(line))

print(f"Loaded {len(papers)} CVPR {year} papers")
```

### Download PDF Files

Each paper has a `download_url` field pointing to the original PDF on CVF Open Access:

```python
import requests
import os

# Create output directory
os.makedirs(f"cvpr_{year}_pdfs", exist_ok=True)

# Download a specific paper
paper = papers[0]
response = requests.get(paper['download_url'])

filename = os.path.basename(paper['pdf_path'])
with open(f"cvpr_{year}_pdfs/{filename}", 'wb') as f:
    f.write(response.content)

print(f"Downloaded: {filename}")

# Download all papers for a year (optional)
for paper in papers:
    if paper.get('download_url'):
        response = requests.get(paper['download_url'])
        filename = os.path.basename(paper['pdf_path'])
        with open(f"cvpr_{year}_pdfs/{filename}", 'wb') as f:
            f.write(response.content)
```

---

## Applications

### PDF Access
- **Direct Download**: Use `download_url` to download PDFs directly from CVF Open Access
- **Full Text Analysis**: Extract text from PDFs for detailed content analysis
- **Figure Extraction**: Extract figures, tables, and equations from papers
- **Layout Analysis**: Analyze paper structure and formatting patterns
- **OCR Processing**: Optical character recognition for scanned documents

### Research Applications
- **Literature Review**: Rapid retrieval of papers in specific domains for comprehensive literature review
- **Trend Analysis**: Analyze research hotspots and development trends in computer vision over the past decade
- **Citation Networks**: Build and analyze paper citation networks based on BibTeX data
- **Knowledge Graphs**: Construct knowledge graphs connecting papers, authors, institutions, and concepts
- **Recommendation Systems**: Build paper recommendation systems based on content similarity

---

## Download

- **Hugging Face**: [datasets/choucsan/CVPR_Papers](https://huggingface.co/datasets/choucsan/CVPR_Papers)

---

## Contact

[choucisan@gmail.com](mailto:choucisan@gmail.com)