Dataset Versions

v3

feather classify no adjusted equalidation

Generated on May 26, 2023

Dataset Split

Train Set 84%
1935Images
Valid Set 11%
251Images
Test Set 5%
106Images

Preprocessing

Static Crop: 6-91% Horizontal Region, 19-62% Vertical Region
Resize: Stretch to 1280x640
Filter Null: Require at least 15% of images to contain annotations.

Augmentations

Outputs per training example: 3
Flip: Horizontal, Vertical
90° Rotate: Clockwise, Counter-Clockwise, Upside Down
Rotation: Between -22° and +22°
Hue: Between -25° and +25°
Brightness: Between -15% and +15%
Blur: Up to 2px
Noise: Up to 5% of pixels