Dataset Versions

v4

2023-06-17 8:52pm

Generated on Jun 17, 2023

Popular Download Formats

Pascal VOC XML
Common XML annotation format for local data munging (pioneered by ImageNet).
PaliGemma
PaliGemma JSONL format used for fine-tuning PaliGemma, Google's open multimodal vision model.
CreateML JSON
CreateML JSON format is used with Apple's CreateML and Turi Create tools.
Other Formats
Choose another format.

Dataset Split

Train Set 87%
27726Images
Valid Set 8%
2664Images
Test Set 4%
1328Images

Preprocessing

Auto-Orient: Applied
Isolate Objects: Applied
Static Crop: 25-76% Horizontal Region, 0-100% Vertical Region
Resize: Stretch to 640x640
Auto-Adjust Contrast: Using Adaptive Equalization
Grayscale: Applied
Tile: 2 rows x 2 columns
Filter Null: Require all images to contain annotations.

Augmentations

Outputs per training example: 3
Flip: Horizontal, Vertical
90° Rotate: Clockwise, Counter-Clockwise, Upside Down
Crop: 0% Minimum Zoom, 10% Maximum Zoom
Rotation: Between -15° and +15°
Shear: ±15° Horizontal, ±15° Vertical
Grayscale: Apply to 25% of images
Hue: Between -25° and +25°
Saturation: Between -25% and +25%
Brightness: Between -25% and +25%
Exposure: Between -25% and +25%
Blur: Up to 3px
Noise: Up to 5% of pixels
Cutout: 3 boxes with 10% size each
Mosaic: Applied
Bounding Box: Flip: Horizontal
Bounding Box: 90° Rotate: Clockwise, Counter-Clockwise
Bounding Box: Crop: 0% Minimum Zoom, 20% Maximum Zoom
Bounding Box: Rotation: Between -15° and +15°
Bounding Box: Shear: ±15° Horizontal, ±15° Vertical
Bounding Box: Brightness: Between -25% and +25%
Bounding Box: Exposure: Between -25% and +25%
Bounding Box: Blur: Up to 2.5px
Bounding Box: Noise: Up to 5% of pixels