Finger Digits 0-5 Dataset

Content:

• 12,000 images of a hand holding up 0-5 fingers.
• Images are 128×128 pixels, thresholded to isolate the hand from the background.
• Most hands are held straight up, lacking sideways and upside-down positions.

Acknowledgements:

• OpenCV library was extensively used.

Inspiration:

• Created due to difficulties with existing datasets in noisy environments and imperfect lighting.

Improvements for New Dataset:

1. Include Diverse Hand Orientations:
   • Capture images with hands held sideways, upside down, and at various angles.
2. Varied Backgrounds and Lighting:
   • Include images with different backgrounds and lighting conditions to increase robustness.
3. Augmentation Techniques:
   • Use rotation, scaling, and flipping to artificially increase dataset diversity.
4. Multiple Skin Tones and Hand Sizes:
   • Ensure diversity in hand types for better generalization.
5. Annotate with Metadata:
   • Include additional annotations like hand orientation and lighting conditions for better training.

This dataset is sourced from Kaggle.