This dataset contains sensitive visual content intended for training Image Guard, Image Filter, and Image Shield models. Due to the sensitive nature of the data, access is gated and requires manual review. Access may be approved for users who provide a valid research, safety, or development purpose.

ImageShield-Guardrail-Realism-60K

ImageShield-Guardrail-Realism-60K is a specialized, curated dataset comprising ~60,000 labeled images focused on evaluating and enforcing visual guardrails across diverse artistic, stylized, and photorealistic distributions. It is tailored for training, fine-tuning, and benchmarking image guardrails, content moderation filters, and safety classifiers (such as ImageShield, Image Guard, and automated visual screening pipelines).

• Curator: prithivMLmods
• Total Samples: 59,997 rows
• Total Size: ~8.67 GB
• Format: Parquet / Image Feature
• Split: Train (60k rows)

Dataset Overview

The dataset provides clean binary annotations classifying visual media into Safe vs. Unsafe categories. It is specifically curated to handle edge cases across varied rendering styles, including stylized, 2D/anime, 3D renders, digital art, and photorealistic representations, with targeted detection of sensitive visual content and safety-risk patterns.

Column	Type	Description
image	Image	Target RGB image across diverse rendering styles
label	ClassLabel	0 = Safe, 1 = Unsafe

Access & Gating Notice

Sensitive Content Warning: This dataset contains sensitive visual content designed solely for AI safety research, image filtering, content moderation, and guardrail training. Access is gated and requires manual verification.

To request access:

1. Sign in to your Hugging Face account.
2. Submit an access request stating your valid research, defense, or safety application purpose.
3. Access will be reviewed and granted accordingly.

How to Use

Using datasets

from datasets import load_dataset

# Load dataset (requires an approved Hugging Face access token)
dataset = load_dataset("prithivMLmods/ImageShield-Guardrail-Realism-60K", split="train")

# Inspect a single sample
sample = dataset[0]
image = sample["image"]
label = sample["label"]

print(f"Label: {'Unsafe' if label == 1 else 'Safe'}")

PyTorch DataLoader Pipeline

import torch
from datasets import load_dataset
from torchvision import transforms
from torch.utils.data import DataLoader

transform = transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
])

def preprocess(batch):
    batch["pixel_values"] = [transform(img.convert("RGB")) for img in batch["image"]]
    return batch

dataset = load_dataset("prithivMLmods/ImageShield-Guardrail-Realism-60K", split="train")
dataset = dataset.with_transform(preprocess)

dataloader = DataLoader(dataset, batch_size=32, shuffle=True)

Intended Uses

• Fine-tuning robust vision backbones (e.g., SigLIP, FastViT, CLIP, ConvNeXt) for visual content safety.
• Input/output safety verification layers for text-to-image diffusion models and multimodal generative pipelines.
• Stylized vs. realistic robustness evaluation in automated content moderation systems.

License

Distributed under the Apache-2.0 License. This dataset is strictly intended for developing defensive AI guardrails, content moderation frameworks, and image safety solutions.