bitmind/bm-image-benchmarks

BitMind Image Benchmarks

This dataset contains AI-generated image samples.

Dataset Structure

Each config represents a batch upload with:

• Parquet files in data/ containing image data and metadata

Loading the Dataset

from datasets import load_dataset

# List available configs (timestamps)
configs = ['split_20250821_110436', 'split_20250821_112432', ...]

# Load specific config
dataset = load_dataset('bitmind/bm-image-benchmarks', 'split_20250821_110436')

# Access data
for sample in dataset['train']:
    print(f"Model: {sample['model_name']}")
    print(f"Label: {sample['label']}")
    # sample['media_image'] contains the PIL Image
    image = sample['media_image']
    image.show()  # Display the image

Processing Images

from datasets import load_dataset
from PIL import Image

# Load dataset (images and metadata)
config = 'split_20250821_110436'  # Use your desired config
dataset = load_dataset('bitmind/bm-image-benchmarks', config)

# Process images with metadata
for sample in dataset['train']:
    image = sample['media_image']  # PIL Image object
    print(f"Model: {sample['model_name']}")
    print(f"Label: {sample['label']}")
    print(f"Hash: {sample['media_hash']}")
    
    # Your image processing code here
    # image.save(f"{sample['media_hash']}.png")