Recognizing Image Style

Abstract

The style of an image plays a significant role in how it is viewed, but style has received little attention in computer vision research. We describe an approach to predicting style of images, and perform a thorough evaluation of different image features for these tasks. We find that features learned in a multi-layer network generally perform besteven when trained with object class (not style) labels. Our large-scale learning methods results in the best published performance on an existing dataset of aesthetic ratings and photographic style annotations. We present two novel datasets: 80K Flickr photographs annotated with 20 curated style labels, and 85K paintings annotated with 25 style/genre labels. Our approach shows excellent classification performance on both datasets. We use the learned classifiers to extend traditional tag-based image search to consider stylistic constraints, and demonstrate cross-dataset understanding of style.

Will It Last? Learning Stable Features for Long-Term Visual Localization

“…ImageNet) and only later, after reducing the network learning rate, try to fine-tune the network to the actual task. Karayev et al [14] state that pretraining a network allows us to reuse the mid-level features learned from the object classification datasets and is generally superior to hand-tuned features.…”

A Learned Representation of Artist-Specific Colourisation

“…There is large body of work on the computational analysis of artworks, while a large portion of this work is concerned with learning characteristics of artists for classification [11,13,26], an increasing body of work is emerging which tries to capture artist-specific characteristics for generative purposes [5,25,4]. This latter type of work, is generally concerned with style transfer (i.e., given a style image S and a content image C produce a single image with style S style and content C content ).…”

Benchmarking data analysis and machine learning applications on the Intel KNL many-core processor

“…Caffe is a deep learning framework developed by Berkeley AI Research Lab [19] and has implemented several published CNN (Convolution Neural Network) models and tutorials such as AlexNet [30,31], LeNet [32,33], GoogleNet [34,35], and an example tutorial for ImageNet data [36]. For this work, we selected an example that finetunes CaffeNet for style recognition on "Flickr style" data [37,38] with using the default single-precision arithmetic. The term, finetuning, means that the new model takes an existing trained model, CaffeNet in this case, adapts the architecture, and resumes training from the already learned model weights [38].…”

“…The MIT SuperCloud has spurred the development of a number of crossecosystem innovations in high performance databases [28,29], database management [30], database federation [31,32,33], data analytics [34], data protection [35], and system monitoring [36,37].…”