{
  "image_size": 256,
  "image_mean": [0.485, 0.456, 0.406],
  "image_std": [0.229, 0.224, 0.225],
  "do_resize": true,
  "do_normalize": true,
  "tta_horizontal_flip": true,
  "input_format": "RGB",
  "output_format": "normalised_xy_in_0_1",
  "notes": "Ultrasound images often come as single-channel. Replicate to 3 channels before normalisation. Horizontal-flip TTA is used at evaluation (average of original and flipped predictions in image space). Output coordinates are normalised to [0, 1] -- multiply by original image width/height to get pixel coordinates."
}