Dataset with labels

All labels per image are stored in a single column as serialized JSON
Each value contains a serialized JSON array of bounding boxes with labels Each value contains one category as string

Configuration

Check to stop training the ResNet backbone.
If a image side is outside this size range, the image will be resized.

Training

Proportion of the sample that goes to the train set. The rest goes to the validation set.
Check to reduce the learning rate when the validation loss is on a plateau.
Number of epochs between two learning rate reduction.
Factor of learning rate reduction.

Data Augmentation

A probability between 0 and 1.
An angle in radians.
A percentage of the image size.
An angle in radians.
A percentage of the image size

GPU

Comma separated list of GPU indexes