We trained a large, deep convolutional neural network to classify the 1.2 million high-resolution images in the ImageNet LSVRC-2010 contest into the 1000 different classes. On the test data, we achieved top-1 and top-5 error rates of 37.5% and 17.0% which is considerably better than the previous state-of-the-art. The neural network, which has 60 million parameters and 650,000 neurons, consists of five convolutional layers, some of which are followed by max-pooling layers, and three fully-connected layers with a final 1000-way softmax. To make training faster, we used non-saturating neurons and a very efficient GPU implementation of the convolution operation. To reduce overriding in the fully-connected layers we employed a recently-developed regularization method called "dropout" that proved to be very effective. We also entered a variant of this model in the ILSVRC-2012 competition and achieved a winning top-5 test error rate of 15.3%, compared to 26.2% achieved by the second-best entry.

/pdf/imagenet-classification-with-deep-convolutional-neural-1190x9a6b9.pdf

ImageNet Classification with Deep Convolutional Neural Networks

In this work we investigate the effect of the convolutional network depth on its accuracy in the large-scale image recognition setting. Our main contribution is a thorough evaluation of networks of increasing depth using an architecture with very small (3x3) convolution filters, which shows that a significant improvement on the prior-art configurations can be achieved by pushing the depth to 16-19 weight layers. These findings were the basis of our ImageNet Challenge 2014 submission, where our team secured the first and the second places in the localisation and classification tracks respectively. We also show that our representations generalise well to other datasets, where they achieve state-of-the-art results. We have made our two best-performing ConvNet models publicly available to facilitate further research on the use of deep visual representations in computer vision.

Very Deep Convolutional Networks for Large-Scale Image Recognition

The explosion of image data on the Internet has the potential to foster more sophisticated and robust models and algorithms to index, retrieve, organize and interact with images and multimedia data. But exactly how such data can be harnessed and organized remains a critical problem. We introduce here a new database called “ImageNet”, a large-scale ontology of images built upon the backbone of the WordNet structure. ImageNet aims to populate the majority of the 80,000 synsets of WordNet with an average of 500-1000 clean and full resolution images. This will result in tens of millions of annotated images organized by the semantic hierarchy of WordNet. This paper offers a detailed analysis of ImageNet in its current state: 12 subtrees with 5247 synsets and 3.2 million images in total. We show that ImageNet is much larger in scale and diversity and much more accurate than the current image datasets. Constructing such a large-scale database is a challenging task. We describe the data collection scheme with Amazon Mechanical Turk. Lastly, we illustrate the usefulness of ImageNet through three simple applications in object recognition, image classification and automatic object clustering. We hope that the scale, accuracy, diversity and hierarchical structure of ImageNet can offer unparalleled opportunities to researchers in the computer vision community and beyond.

/pdf/imagenet-a-large-scale-hierarchical-image-database-1e8f663s1h.pdf

ImageNet: A large-scale hierarchical image database

We trained a large, deep convolutional neural network to classify the 1.2 million high-resolution images in the ImageNet LSVRC-2010 contest into the 1000 different classes. On the test data, we achieved top-1 and top-5 error rates of 37.5% and 17.0%, respectively, which is considerably better than the previous state-of-the-art. The neural network, which has 60 million parameters and 650,000 neurons, consists of five convolutional layers, some of which are followed by max-pooling layers, and three fully connected layers with a final 1000-way softmax. To make training faster, we used non-saturating neurons and a very efficient GPU implementation of the convolution operation. To reduce overfitting in the fully connected layers we employed a recently developed regularization method called "dropout" that proved to be very effective. We also entered a variant of this model in the ILSVRC-2012 competition and achieved a winning top-5 test error rate of 15.3%, compared to 26.2% achieved by the second-best entry.

/pdf/imagenet-classification-with-deep-convolutional-neural-3kv7rezkq1.pdf

ImageNet classification with deep convolutional neural networks

We introduce a challenging set of 256 object categories containing a total of 30607 images The original Caltech-101 [1] was collected by choosing a set of object categories, downloading examples from Google Images and then manually screening out all images that did not fit the category Caltech-256 is collected in a similar manner with several improvements: a) the number of categories is more than doubled, b) the minimum number of images in any category is increased from 31 to 80, c) artifacts due to image rotation are avoided and d) a new and larger clutter category is introduced for testing background rejection We suggest several testing paradigms to measure classification performance, then benchmark the dataset using two simple metrics as well as a state-of-the-art spatial pyramid matching [2] algorithm Finally we use the clutter category to train an interest detector which rejects uninformative background regions

Caltech-256 Object Category Dataset

The computational complexity of current visual categorization algorithms scales linearly at best with the number of categories. The goal of classifying simultaneously Ncat = 104 - 105 visual categories requires sub-linear classification costs. We explore algorithms for automatically building classification trees which have, in principle, logNcat complexity. We find that a greedy algorithm that recursively splits the set of categories into the two minimally confused subsets achieves 5-20 fold speedups at a small cost in classification performance. Our approach is independent of the specific classification algorithm used. A welcome by-product of our algorithm is a very reasonable taxonomy of the Caltech-256 dataset.

http://www.vision.caltech.edu/publications/griffinperonaCVPR2008.pdf

Learning and using taxonomies for fast visual categorization

We have observed the linear polarization of 450 µm continuum emission from the Galactic center, using a new polarimetric detector system that is operated on a 2 m telescope at the South Pole. The resulting polarization map extends � 170 pc along the Galactic plane and � 30 pc in Galactic latitude, and thus covers a significant fraction of the central molecular zone. Our map shows that this region is permeated by large-scale toroidal magnetic fields. We consider our results together with radio observations that show evidence for poloidal fields in the Galactic center, and with Faraday rotation observations. We compare all of these observations with the predictions of a magnetodynamic model for the Galactic center that was proposed in order to explain the Galactic Center Radio Lobe as a magnetically driven gas outflow. We conclude that the observations are basically consistent with the model.

First Results from the Submillimeter Polarimeter for Antarctic Remote Observations: Evidence of Large-Scale Toroidal Magnetic Fields in the Galactic Center

We have observed the linear polarization of 450 micron continuum emission from the Galactic center, using a new polarimetric detector system that is operated on a 2 m telescope at the South Pole. The resulting polarization map extends ~ 170 pc along the Galactic plane and ~ 30 pc in Galactic latitude, and thus covers a significant fraction of the central molecular zone. Our map shows that this region is permeated by large-scale toroidal magnetic fields. We consider our results together with radio observations that show evidence for poloidal fields in the Galactic center, and with Faraday rotation observations. We compare all of these observations with the predictions of a magnetodynamic model for the Galactic center that was proposed in order to explain the Galactic Center Radio Lobe as a magnetically driven gas outflow. We conclude that the observations are basically consistent with the model.

First Results from SPARO: Evidence for Large-Scale Toroidal Magnetic Fields in the Galactic Center

We present results from the Austral Winter 2003 observing campaign of SPARO, a 450 μm polarimeter used with a 2 m telescope at the South Pole. We mapped large-scale magnetic fields in four GMCs in the Galactic disk: NGC 6334, the Carina Nebula, G333.6-0.2, and G331.5-0.1. We find a statistically significant correlation of the inferred field directions with the orientation of the Galactic plane. Specifically, three of the four GMCs (NGC 6334 is the exception) have mean field directions that are within 15° of the plane. The simplest interpretation is that the field direction tends to be preserved during the process of GMC formation. We have also carried out an analysis of published optical polarimetry data. For the closest of the SPARO GMCs, NGC 6334, we can compare the field direction in the cloud as measured by SPARO with the field direction in a larger region surrounding the cloud, as determined from optical polarimetry. For purposes of comparison, we also use optical polarimetry to determine field directions for 9-10 other regions of similar size. We find that the region surrounding NGC 6334 is an outlier in the distribution of field directions determined from optical polarimetry, just as the NGC 6334 cloud is an outlier in the distribution of cloud field directions determined by SPARO. In both cases the field direction corresponding to NGC 6334 is rotated away from the direction of the plane by a large angle. This finding is consistent with our suggestion that field direction tends to be preserved during GMC formation. Finally, by comparing the disorder in our magnetic field maps with the disorder seen in magnetic field maps derived from MHD turbulence simulations, we conclude that the magnetic energy density in our clouds is comparable to the turbulent energy density.

/pdf/results-of-sparo-2003-mapping-magnetic-fields-in-giant-4acgu6d0l2.pdf

G. S. Griffin

Papers

Caltech-256 Object Category Dataset

Learning and using taxonomies for fast visual categorization

First Results from the Submillimeter Polarimeter for Antarctic Remote Observations: Evidence of Large-Scale Toroidal Magnetic Fields in the Galactic Center

First Results from SPARO: Evidence for Large-Scale Toroidal Magnetic Fields in the Galactic Center

Results of SPARO 2003: Mapping Magnetic Fields in Giant Molecular Clouds