skip connections

This post dives into OSNet, a lightweight yet powerful network designed for person re-identification. It introduces the concept of omni-scale feature learning, where multi-scale features are captured and fused dynamically inside a single residual block. From architecture to performance, we break down what makes OSNet both elegant and effective.

ResNet changed deep learning forever. Before its arrival, training very deep networks was nearly impossible: deep models did not learn! But with a simple yet powerful trick ResNet shattered these barriers. Dive in to discover how ResNet works.