Dataset Versions

v3

2024-02-22 6:06pm

Generated on Feb 22, 2024

Dataset Split

Train Set 92%
4000Images
Valid Set 3%
132Images
Test Set 5%
200Images

Preprocessing

Auto-Orient: Applied
Resize: Fit within 640x640
Modify Classes: 0 remapped, 21 dropped
Filter Null: Require all images to contain annotations.

Augmentations

Outputs per training example: 2
Brightness: Between -10% and +10%
Exposure: Between -3% and +3%
Blur: Up to 0.4px