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

We propose a novel approach to unsupervised facial image alignment. Differently from previous approaches, that are confined to affine transformations on either the entire face or separate patches, we extract a nonrigid mapping between facial images. Based on a regularized face model, we frame unsupervised face alignment into the Lucas-Kanade image registration approach. We propose a robust optimization scheme to handle appearance variations. The method is fully automatic and can cope with pose variations and expressions, all in an unsupervised manner. Experiments on a large set of images showed that the approach is effective.

/pdf/unsupervised-face-alignment-by-robust-nonrigid-mapping-1u3uwy4ugt.pdf

Unsupervised face alignment by robust nonrigid mapping

. In this paper, we present a system that was developed for the European Space Agency (ESA) for the support of planetary exploration. The system that is sent to the planetary surface consists of a rover and a lander. The lander contains a stereo head equipped with a pan-tilt mechanism. This vision system is used both for modeling the terrain and for localization of the rover. Both tasks are necessary for the navigation of the rover. Due to the stress that occurs during the flight, a recalibration of the stereo-vision system is required once it is deployed on the planet. Practical limitations make it unfeasible to use a known calibration pattern for this purpose; therefore, a new calibration procedure had to be developed that could work on images of the planetary environment. This automatic procedure recovers the relative orientation of the cameras and the pan and tilt axes, as well as the exterior orientation for all the images. The same images are subsequently used to reconstruct the 3-D structure of the terrain. For this purpose, a dense stereo-matching algorithm is used that (after rectification) computes a disparity map. Finally, all the disparity maps are merged into a single digital terrain model. In this paper, a simple and elegant procedure is proposed that achieves that goal. The fact that the same images can be used for both calibration and 3-D reconstruction is important, since, in general, the communication bandwidth is very limited. In addition to navigation and path planning, the 3-D model of the terrain is also used for virtual-reality simulations of the mission, wherein the model is texture mapped with the original images. The system has been implemented, and the first tests on the ESA planetary terrain testbed were successful.

A stereo-vision system for support of planetary surface exploration

Hough transform based methods for object detection work by allowing image features to vote for the location of the object. While this representation allows for parts observed in different training instances to support a single object hypothesis, it also produces false positives by accumulating votes that are consistent in location but inconsistent in other properties like pose, color, shape or type. In this work, we propose to augment the Hough transform with latent variables in order to enforce consistency among votes. To this end, only votes that agree on the assignment of the latent variable are allowed to support a single hypothesis. For training a Latent Hough Transform (LHT) model, we propose a learning scheme that exploits the linearity of the Hough transform based methods. Our experiments on two datasets including the challenging PASCAL VOC 2007 benchmark show that our method outperforms traditional Hough transform based methods leading to state-of-the-art performance on some categories.

/pdf/latent-hough-transform-for-object-detection-m583fwr52p.pdf

Latent hough transform for object detection

In this work, we present a method for action localization and recognition using an exemplar-based approach. It starts from local dense yet scale-invariant spatio-temporal features. The most discriminative visual words are selected and used to cast bounding box hypotheses, which are then verified and further grouped into the final detections. To the best of our knowledge, we are the first to extend the exemplar-based approach using local features into the spatio-temporal domain. This allows us to avoid the problems that typically plague sliding window-based approaches - in particular the exhaustive search over spatial coordinates, time, and spatial as well as temporal scales. We report state-ofthe-art results on challenging datasets, extracted from real movies, for both classification and localization.

/pdf/exemplar-based-action-recognition-in-video-3cm498s51q.pdf

Exemplar-based Action Recognition in Video

Being able to learn dense semantic representations of images without supervision is an important problem in computer vision. However, despite its significance, this problem remains rather unexplored, with a few exceptions that considered unsupervised semantic segmentation on small-scale datasets with a narrow visual domain. In this paper, we make a first attempt to tackle the problem on datasets that have been traditionally utilized for the supervised case. To achieve this, we introduce a two-step framework that adopts a predetermined mid-level prior in a contrastive optimization objective to learn pixel embeddings. This marks a large deviation from existing works that relied on proxy tasks or end-to-end clustering. Additionally, we argue about the importance of having a prior that contains information about objects, or their parts, and discuss several possibilities to obtain such a prior in an unsupervised manner. 
Experimental evaluation shows that our method comes with key advantages over existing works. First, the learned pixel embeddings can be directly clustered in semantic groups using K-Means on PASCAL. Under the fully unsupervised setting, there is no precedent in solving the semantic segmentation task on such a challenging benchmark. Second, our representations can improve over strong baselines when transferred to new datasets, e.g. COCO and DAVIS. The code is available.

Luc Van Gool

Papers

Unsupervised face alignment by robust nonrigid mapping

A stereo-vision system for support of planetary surface exploration

Latent hough transform for object detection

Exemplar-based Action Recognition in Video

Unsupervised Semantic Segmentation by Contrasting Object Mask Proposals