Hi,
Does anyone know how to implement data augmentation for multiple coordinate regression? Let’s say, the output is a 4-dimensional vector [x1, y1, x2, y2] of real numbers representing two coordinates (x1, y1), (x2, y2).
In the past year, I have implemented pupillary distance measurement in TensorFlow. Now I would like to try to evaluate U-Net with fast.ai.
For TensorFlow, I have implemented the augmentation logic from scratch to recalculate coordinates in case of padding, translating, and flipping. If it is possible, I would like to take advantage of possibly existing augmentation logic implemented in fast.ai.