Deeper neural networks are more difficult to train. We present a residual learning framework to ease the training of networks that are substantially deeper than those used previously. We explicitly reformulate the layers as learning residual functions with reference to the layer inputs, instead of learning unreferenced functions. We provide comprehensive empirical evidence showing that these residual networks are easier to optimize, and can gain accuracy from considerably increased depth. On the ImageNet dataset we evaluate residual nets with a depth of up to 152 layers—8× deeper than VGG nets [40] but still having lower complexity. An ensemble of these residual nets achieves 3.57% error on the ImageNet test set. This result won the 1st place on the ILSVRC 2015 classification task. We also present analysis on CIFAR-10 with 100 and 1000 layers. The depth of representations is of central importance for many visual recognition tasks. Solely due to our extremely deep representations, we obtain a 28% relative improvement on the COCO object detection dataset. Deep residual nets are foundations of our submissions to ILSVRC & COCO 2015 competitions1, where we also won the 1st places on the tasks of ImageNet detection, ImageNet localization, COCO detection, and COCO segmentation.

/pdf/deep-residual-learning-for-image-recognition-b75jg61qrq.pdf

Deep Residual Learning for Image Recognition

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

Deeper neural networks are more difficult to train. We present a residual learning framework to ease the training of networks that are substantially deeper than those used previously. We explicitly reformulate the layers as learning residual functions with reference to the layer inputs, instead of learning unreferenced functions. We provide comprehensive empirical evidence showing that these residual networks are easier to optimize, and can gain accuracy from considerably increased depth. On the ImageNet dataset we evaluate residual nets with a depth of up to 152 layers---8x deeper than VGG nets but still having lower complexity. An ensemble of these residual nets achieves 3.57% error on the ImageNet test set. This result won the 1st place on the ILSVRC 2015 classification task. We also present analysis on CIFAR-10 with 100 and 1000 layers. 
The depth of representations is of central importance for many visual recognition tasks. Solely due to our extremely deep representations, we obtain a 28% relative improvement on the COCO object detection dataset. Deep residual nets are foundations of our submissions to ILSVRC & COCO 2015 competitions, where we also won the 1st places on the tasks of ImageNet detection, ImageNet localization, COCO detection, and COCO segmentation.

We propose a deep convolutional neural network architecture codenamed Inception that achieves the new state of the art for classification and detection in the ImageNet Large-Scale Visual Recognition Challenge 2014 (ILSVRC14). The main hallmark of this architecture is the improved utilization of the computing resources inside the network. By a carefully crafted design, we increased the depth and width of the network while keeping the computational budget constant. To optimize quality, the architectural decisions were based on the Hebbian principle and the intuition of multi-scale processing. One particular incarnation used in our submission for ILSVRC14 is called GoogLeNet, a 22 layers deep network, the quality of which is assessed in the context of classification and detection.

/pdf/going-deeper-with-convolutions-1yobw2o2ds.pdf

Going deeper with convolutions

3D city modeling is a very demanding task. It suffers from the same problems as general bottom-up 3D acquisition processes. Whatever the 3D acquisition system, there are always objects and surfaces that yield meager performance. Often the problems are fundamental and cannot be resolved through more sophisticated bottom-up processing. If however, one knows what kind of objects are to be modeled – e.g. buildings – strong prior knowledge can be invoked. Inverse procedural modeling is a case in point. It yields more compact and more realistic models, yet requires a grammar to be created and then activated for the style at hand. The paper discusses style classification and the (initial) use of more generic architectural guidelines as ways to mitigate the problems with procedural grammars.

Towards semantic city models

A stereo rig can be calibrated using a calibration grid, but recent work demonstrated the possibility of auto-calibration. There remain two important limitations, however. First, the focal lengths of the cameras should remain fixed, thereby excluding zooming or focusing. Second, the stereo rig must not purely translate, which however is the most natural type of motion. This also implies that these methods collapse when the motion comes close to being a translation.

Euclidean 3D reconstruction from stereo sequences with variable focal lengths

An automatic method for acquiring 3D models from photographs : applications to an archaeological site

This paper tackles the task of estimating the topology of filamentary networks such as retinal vessels and road networks. Building on top of a global model that performs a dense semantical classification of the pixels of the image, we design a Convolutional Neural Network (CNN) that predicts the local connectivity between the central pixel of an input patch and its border points. By iterating this local connectivity we sweep the whole image and infer the global topology of the filamentary network, inspired by a human delineating a complex network with the tip of their finger. 
We perform an extensive and comprehensive qualitative and quantitative evaluation on two tasks: retinal veins and arteries topology extraction and road network estimation. In both cases, represented by two publicly available datasets (DRIVE and Massachusetts Roads), we show superior performance to very strong baselines.

Iterative Deep Learning for Network Topology Extraction

Despite recent advances in computer vision and large-scale indexing techniques, automatic geo-localization of images and videos remains a challenging task. The majority of existing computer vision solutions for geo-localization are limited to highly-visited urban regions for which a significant amount of geo-tagged imagery is available, and therefore, do not scale well to large and ordinary geo-spatial regions. In this chapter, we provide an overview of the major research themes in visual geo-localization, investigate the challenges, and point to problem areas that will benefit from common synthesis of perspectives from these research themes. In particular, we discuss how the availability of web-scale geo-referenced data affects visual geo-localization, what role semantic information plays in this problem, and how precise localization can be achieved using large-scale textured (RGB) and untextured (non-RGB) 3D models. We also introduce a few real-world applications which became feasible as a result of the capability of estimating an image’s geo-location. We conclude this chapter by providing an overview of the emerging trends in visual geo-localization and a summary of the rest of the chapters of the book.

Luc Van Gool

Papers

Towards semantic city models

Euclidean 3D reconstruction from stereo sequences with variable focal lengths

An automatic method for acquiring 3D models from photographs : applications to an archaeological site

Iterative Deep Learning for Network Topology Extraction

Introduction to Large-Scale Visual Geo-localization