#edge-detection

Question 1: Select all that apply

mediumAI-checkedConvolutional Neural Networks

A $6 \times 6$ image X has ones on and below the main diagonal and zeros above it (X = torch.tril(torch.ones(6, 6))), so it contains a diagonal edge. It is cross-correlated with the finite-difference kernel K = [[1, -1]], which computes $x_{i,j} - x_{i,j+1}$ . Select all that apply.

The output has the same $6 \times 6$ shape as X1
The output is zero everywhere, because K only detects vertical edges2
Transposing the image with X.T makes the response of K vanish, just as transposing a vertical edge does3
The output is nonzero along the diagonal, one entry per row except the last4
The transposed kernel [[1], [-1]], which takes vertical differences, also responds to this edge5

Timed test