Mafu-Labs/datadive-360k

datadive

[image]

Datadive is a large-scale color detection dataset containing 360,000 images. Each image is 128x128 pixels and generated with a variety of colors, shapes, and realistic background effects. It’s ideal for testing and training models related to color recognition, object detection, and image classification.

Dataset Details

• —Number of images: 360,000
• —Image size: 128 x 128 pixels
• —Categories / Colors: red, green, blue, yellow, cyan, orange, black, white, gray, lime, navy, brown
• —Format: PNG
• —Purpose: Color detection, classification, and computer vision experiments

Features

• —Randomized backgrounds: solid, gradient, noise, corruption, and perlin noise.
• —Random shapes: circles, rectangles, and triangles.
• —Lighting, shadow, and noise effects applied for realism.
• —Optional texture overlays for variety (stored in the textures/ folder).

Structure

datadive/
├── banner.png
├── black/
│   ├── black_1.png
│   ├── black_2.png
│   └── ...
├── blue/
│   ├── blue_1.png
│   └── ...
├── ...
├── red/
├── white/
└── yellow/

Each color has its own folder containing 30,000 images.

Usage

Using Hugging Face datasets (if stored as Parquet)

from datasets import load_dataset
import io
from PIL import Image

dataset = load_dataset("Mafu-Labs/datadive", split="train")

for example in dataset:
    img_bytes = example["image"]
    img = Image.open(io.BytesIO(img_bytes))
    img.show()

Direct file access

from PIL import Image
img = Image.open("datadive/blue/blue_1.png")
img.show()

License

This dataset is released under the Apache 2.0 License.