#semantic-segmentation

Question 1: Select all that apply

You reuse image-classification augmentations to train a semantic segmentation model, where every pixel has a class label. Which statements are correct? Select all that apply.

Random cropping cannot be used, because a crop would cut objects at its border1
A random horizontal flip must be applied to the label map too, with the same random draw2
Resizing a random crop with bilinear interpolation is just as safe for the label map as for the image3
Color jitter (brightness, contrast, hue) can be applied to the input image alone4

#semantic-segmentation

Timed test