Utilities to process your images before performing an inference.
ResamplingWithOriginalRatio(img, required_size, sample)¶
Resizes the image to maintain the original aspect ratio by adding pixel padding where needed.
For example, if your model’s input tensor requires a square image but your image is landscape (and you don’t want to reshape the image to fit), pass this function your image and the required square dimensions, and it returns a square version by adding the necessary amount of black pixels on the bottom-side only. If the original image is portrait, it adds black pixels on the right-side only.
- img (
PIL.Image) – The image to resize.
- required_size (list) – The pixel width and height [x, y] that your model requires for input.
- sample (int) – A resampling filter for image resizing.
This can be one of
PIL.Image.LANCZOS. See Pillow filters.
A 2-tuple with a
PIL.Imageobject for the resized image, and a tuple of floats representing the aspect ratio difference between the original image and the returned image (x delta-ratio, y delta-ratio).
- img (
Is this content helpful?