OnionNet: Sharing Features in Cascaded Deep Classifiers

TL;DR

OnionNet introduces a feature-sharing cascade architecture for deep neural networks that accelerates evaluation in retrieval tasks by sharing intermediate features, maintaining high accuracy with reduced computation.

Contribution

The paper presents a novel cascade model with shared features, trained end-to-end, that improves evaluation speed while preserving accuracy in deep neural network applications.

Findings

Operates significantly faster than monolithic networks.

Maintains comparable accuracy with marginal decrease.

Validated on three diverse applications.

Abstract

The focus of our work is speeding up evaluation of deep neural networks in retrieval scenarios, where conventional architectures may spend too much time on negative examples. We propose to replace a monolithic network with our novel cascade of feature-sharing deep classifiers, called OnionNet, where subsequent stages may add both new layers as well as new feature channels to the previous ones. Importantly, intermediate feature maps are shared among classifiers, preventing them from the necessity of being recomputed. To accomplish this, the model is trained end-to-end in a principled way under a joint loss. We validate our approach in theory and on a synthetic benchmark. As a result demonstrated in three applications (patch matching, object detection, and image retrieval), our cascade can operate significantly faster than both monolithic networks and traditional cascades without sharing at the cost of marginal decrease in precision.