Multi-scale recognition with DAG-CNNs

TL;DR

This paper introduces DAG-CNNs, a multi-scale CNN architecture that extracts features from multiple layers to improve image classification, achieving state-of-the-art results on scene benchmarks.

Contribution

The paper proposes DAG-CNNs, a novel multi-scale CNN structure that leverages features from multiple layers for enhanced classification performance.

Findings

Achieves state-of-the-art accuracy on SUN397, MIT67, and Scene15 datasets.

Multi-scale features outperform single-layer features in classification tasks.

Off-the-shelf multi-scale features perform competitively without fine-tuning.

Abstract

We explore multi-scale convolutional neural nets (CNNs) for image classification. Contemporary approaches extract features from a single output layer. By extracting features from multiple layers, one can simultaneously reason about high, mid, and low-level features during classification. The resulting multi-scale architecture can itself be seen as a feed-forward model that is structured as a directed acyclic graph (DAG-CNNs). We use DAG-CNNs to learn a set of multiscale features that can be effectively shared between coarse and fine-grained classification tasks. While fine-tuning such models helps performance, we show that even "off-the-self" multiscale features perform quite well. We present extensive analysis and demonstrate state-of-the-art classification performance on three standard scene benchmarks (SUN397, MIT67, and Scene15). In terms of the heavily benchmarked MIT67 and Scene15 datasets, our results reduce the lowest previously-reported error by 23.9% and 9.5%, respectively.