rezzzq/RSCD-1million

RSCD: Road Surface Condition Dataset

Dataset Description

The Road Surface Condition Dataset (RSCD) is a large-scale image dataset containing over 1 million images for road surface condition classification. This dataset is designed for training computer vision models to identify and classify various road surface types, moisture conditions, and damage severity levels.

Dataset Summary

• —Total Images: ~1,028,000 images
• —Image Format: JPG
• —Use Cases:
• —Road condition monitoring
• —Autonomous driving systems
• —Infrastructure maintenance
• —Weather-aware navigation

Dataset Structure

The dataset is organized into three splits:

RSCD dataset-1million/
├── train/           # Training set (~900k images, organized by category folders)
├── test_50k/        # Test set (50k images, flat structure)
└── vali_20k/        # Validation set (20k images, flat structure)

Training Set Organization

The training data is organized into category-specific folders:

Surface Types:

• —Asphalt
• —Concrete
• —Gravel
• —Mud

Moisture Conditions:

• —Dry
• —Wet
• —Water (standing water/puddles)

Severity Levels (for asphalt and concrete):

• —Smooth (no damage)
• —Slight (minor damage)
• —Severe (significant damage)

Special Conditions:

• —Fresh snow
• —Ice
• —Melted snow

Example Categories

• —dry_asphalt_smooth - Dry asphalt with no damage
• —wet_concrete_severe - Wet concrete with severe damage
• —water_asphalt_slight - Asphalt with standing water and slight damage
• —dry_gravel - Dry gravel road
• —ice - Icy road surface

File Naming Convention

Test and validation images follow the pattern:

{timestamp}-{moisture}-{surface}-{severity}.jpg

Example: 2022012523413511-wet-asphalt-smooth.jpg

Label Distribution

The dataset contains 27 distinct categories covering various combinations of:

• —4 surface types (asphalt, concrete, gravel, mud)
• —3 moisture conditions (dry, wet, water)
• —3 severity levels (smooth, slight, severe)
• —3 weather conditions (freshsnow, ice, meltedsnow)

Usage

Loading the Dataset

from datasets import load_dataset

# Load the full dataset
dataset = load_dataset("rezzzq/RSCD-1million")

# Access specific splits
train_data = dataset['train']
test_data = dataset['test']
validation_data = dataset['validation']

Use with PyTorch

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

# Load dataset
dataset = load_dataset("rezzzq/RSCD-1million", split="train")

# Define transforms
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])
])

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

Applications

This dataset can be used for:

1. 1.Road Condition Classification: Train models to automatically identify road surface conditions
2. 2.Autonomous Vehicle Systems: Enable vehicles to adapt driving behavior based on road conditions
3. 3.Infrastructure Monitoring: Automated detection of road damage for maintenance prioritization
4. 4.Weather Impact Assessment: Study effects of weather on road surface conditions
5. 5.Multi-task Learning: Train models to simultaneously predict surface type, moisture, and damage level

Recommended Label Merging

For simplified classification tasks, you may want to merge labels:

Surface + Severity Only (ignore moisture)

Merge moisture variants to focus on road type and damage:

• —asphalt-smooth (dry/wet/water variants combined)
• —asphalt-slight
• —asphalt-severe
• —concrete-smooth
• —concrete-slight
• —concrete-severe
• —gravel
• —mud

This reduces classes from 27 to ~12 while increasing samples per class 3x.

Citation

If you use this dataset in your research, please cite:

@dataset{rscd_1million,
  title={RSCD: Road Surface Condition Dataset},
  author={rezzzq},
  year={2022},
  publisher={HuggingFace},
  url={https://huggingface.co/datasets/rezzzq/RSCD-1million}
}

License

MIT License

Dataset Statistics

• —Training Images: ~900,000
• —Validation Images: ~20,000
• —Test Images: ~50,000
• —Total Size: ~1,028,000 images
• —Categories: 27 distinct road condition combinations

Contact

For questions or issues regarding this dataset, please open an issue on the dataset repository.