PNC-07/PNC-TechLabs.IOC1_Dataset_v2

IOC1 Dataset v2

A standardized object-centric computer vision dataset designed for AI Safety, Computer Vision, and Machine Learning research.

---

Overview

IOC1 Dataset v2 is a curated object-centric dataset containing synchronized image assets and rich metadata for computer vision research.

Each sample includes:

• Original Image
• Segmentation Mask
• Object-only Image
• Background-only Image
• Metadata

The dataset is released in two formats:

• Original Dataset – Human-readable, inspection and debugging
• WebDataset – High-performance streaming for large-scale training

---

Designed For

• AI Safety Research
• Computer Vision
• Machine Learning
• Object Classification
• Segmentation
• Context Understanding
• Vision Foundation Models

---

Features

• Object-centric samples
• Binary segmentation masks
• Object/background separation
• Rich metadata
• Train / Validation / Test splits
• WebDataset support
• Streaming-ready
• Reproducible dataset structure
• Dataset validation metadata

---

Repository Structure

IOC1_Dataset_v2/
│
├── IOC1_Dataset_v2_Original/
│   ├── Data/
│   └── Metadata/
│
├── IOC1_Dataset_v2_WebDataset/
│   ├── Data/
│   ├── Metadata/
│   └── Work/
│
├── README.md
├── LICENSE
└── CITATION.cff

---

Quick Start

Kaggle

from kaggle.api.kaggle_api_extended import KaggleApi

api = KaggleApi()
api.authenticate()

api.dataset_download_files(
    "YOUR_USERNAME/IOC1_Dataset_v2",
    path="./dataset",
    unzip=True
)

Hugging Face

from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="YOUR_ORG/IOC1_Dataset_v2",
    repo_type="dataset"
)

---

Dataset Components

Component	Description
Image	Original RGB image
Mask	Binary segmentation mask
Object	Foreground-only image
Background	Background-only image
Metadata	JSON/JSONL sample information

---

Intended Usage

Recommended for:

• Model Training
• Benchmarking
• Segmentation
• Representation Learning
• AI Safety Research

---

Dataset Formats

Format	Purpose
Original	Inspection & Analysis
WebDataset	Training & Streaming

---

Documentation

The remainder of this README is the complete technical reference manual covering:

• Dataset structure
• Metadata specification
• File formats
• Loading examples
• Validation
• Best practices
• Licensing
• Citation

---

Citation

If you use this dataset, please cite and credit PNC-TechLabs.

See:

• CITATION.cff
• License section

---

License

Released under the PNC-TechLabs Attribution License (PTAL-1.0).

Commercial use, research, redistribution, and derivative works are permitted, provided attribution to PNC-TechLabs is maintained.