Return of the Devil in the Details: Delving Deep into Convolutional Nets

Abstract

Abstract-The latest generation of Convolutional Neural Networks (CNN) have achieved impressive results in challenging benchmarks on image recognition and object detection, significantly raising the interest of the community in these methods. Nevertheless, it is still unclear how different CNN methods compare with each other and with previous state-of-the-art shallow representations such as the Bag-of-Visual-Words and the Improved Fisher Vector. This paper conducts a rigorous evaluation of these new techniques, exploring different deep architectures and comparing them on a common ground, identifying and disclosing important implementation details. We identify several useful properties of CNN-based representations, including the fact that the dimensionality of the CNN output layer can be reduced significantly without having an adverse effect on performance. We also identify aspects of deep and shallow methods that can be successfully shared. In particular, we show that the data augmentation techniques commonly applied to CNN-based methods can also be applied to shallow methods, and result in an analogous performance boost. Source code and models to reproduce the experiments in the paper is made publicly available.

A multimodal feature learning approach for sentiment analysis of social network multimedia

“…Popular image representations are the Visual Bag-Of-Words Model [19,26,29], Fisher Vector [35] and its improved version [1,36]. However, as shown recently in [11,24], neural network based ) and share the same polarity models have been shown to widely outperform these previous models. So, to fit with the CBOW representation discussed in the previous section, we choose to exploit the images by using a representation similar to the one used for the textual information, i.e.…”

Matching-range-constrained real-time loop closure detection with CNNs features

“…A comprehensive evaluation further demonstrates the advantages of deep CNNs features with respect to shallow handcrafted feature for image classification [22]. The advantage is that researchers will be free from mastering the knowledge of specific domains and the CNNs architecture can be used for many different domains, especially in visual systems with minor changes.…”

Image-Mediated Learning for Zero-Shot Cross-Lingual Document Retrieval

“…CNN has improved dramatically over the last few years, and many new powerful pre-trained networks are currently available. We compared three different features extracted from GoogLeNet (Szegedy et al, 2014), VGG 16 layers (Chatfield et al, 2014), and CaffeNet (Jia et al, 2014;Krizhevsky et al, 2012). Additionally, we tested the Fisher Vector (Perronnin et al, 2010), which was the standard hand-crafted image feature before deep learning.…”