All Stories

Understanding Padding in Convolutional Neural Networks

Why convolution shrinks its input and under-uses the border, the p = (f-1)/2 rule that fixes both, and why filter sizes are almost always odd.

• cnn
• computer-vision
• coursera
• deep-learning
• Written in EN

Daniel Du
21 Feb 2022 · 4 min read

Edge Detection and the Convolution Operation

What the convolution operation actually computes, worked through a vertical edge detector, and why deep learning learns the filter instead of designing it.

• cnn
• computer-vision
• coursera
• deep-learning
• Written in EN

Daniel Du
20 Feb 2022 · 5 min read