Luc Van Gool

Bio: Luc Van Gool is an academic researcher from Katholieke Universiteit Leuven. The author has contributed to research in topics: Computer science & Object detection. The author has an hindex of 133, co-authored 1307 publications receiving 107743 citations. Previous affiliations of Luc Van Gool include Microsoft & ETH Zurich.

Towards a combined use of IR, UV and 3D-Imaging for the study of small inscribed and illuminated artefacts

Abstract: When heritage objects are being transformed into digital representations, the loss of information is inevitable. The challenge lies in developing integrated systems able to minimize this loss and bring together as many different kinds of recordable characteristics as possible of one and the same object. This contribution presents an approach that combines the detection of colour, surface shape and the reflective characteristics of surfaces by using a selection of IR, Red, Green, Blue and UV light spectra and applying them on 3D models. A multispectral, multi-directional, portable and dome-shaped recording tool has been developed to this end. With the associated software, virtual relighting and enhancements can be applied in an interactive manner based on the principles of photometric stereo: this allows alternating in real time between computations with IR, R, G, B and UV light spectra. Throughout the testing phases, this non-invasive registration and documentation technique has been applied to monitor and study a vast number of heritage objects, varying from 19 c. BCE Egyptian inscribed clay figurines to medieval illuminations.

Monkeys - a software architecture for viroom - low-cost multicamera system

Abstract: This paper presents a software architecture for a software-synchronized multicamera setup. The software allows consistent multiimage acquisition, image processing and decision making. It consists of several programs that perform certain tasks and communicate with each other through shared memory space or TCP/IP packets. The software system can easily accommodate different number of FireWire digital cameras and networked computers. Minimal hardware requirement is the main advantage of the system which makes it flexible and transportable. The functionality of the system is demonstrated on a simple telepresence application.

DLDR: Deep Linear Discriminative Retrieval for Cultural Event Classification from a Single Image

Abstract: In this paper we tackle the classification of cultural events from a single image with a deep learning based method. We use convolutional neural networks (CNNs) with VGG-16 architecture [17], pretrained on ImageNet or the Places205 dataset for image classification, and fine-tuned on cultural events data. CNN features are robustly extracted at 4 different layers in each image. At each layer Linear Discriminant Analysis (LDA) is employed for discriminative dimensionality reduction. An image is represented by the concatenated LDA-projected features from all layers or by the concatenation of CNN pooled features at each layer. The classification is then performed through the Iterative Nearest Neighbors-based Classifier (INNC) [20]. Classification scores are obtained for different image representation setups at train and test. The average of the scores is the output of our deep linear discriminative retrieval (DLDR) system. With 0.80 mean average precision (mAP) DLDR is a top entry for the ChaLearn LAP 2015 cultural event recognition challenge.

Lens Distortion Recovery for Accurate Sequential Structure and Motion Recovery

Abstract: Lens distortions in off-the-shelf or wide-angle cameras block the road to high accuracy Structure and Motion Recovery (SMR) from video sequences. Neglecting lens distortions introduces a systematic error buildup which causes recovered structure and motion to bend and inhibits turntable or other loop sequences to close perfectly. Locking back onto previously reconstructed structure can become impossible due to the large drift caused by the error buildup. Bundle adjustments are widely used to perform an ultimate post-minimization of the total reprojection error. However, the initial recovered structure and motion needs to be close to optimal to avoid local minima. We found that bundle adjustments cannot remedy the error buildup caused by ignoring lens distortions. The classical approach to distortion removal involves a preliminary distortion estimation using a calibration pattern, known geometric properties of perspective projections or only 2D feature correspondences. Often the distortion is assumed constant during camera usage and removed from the images before applying SMR algorithms. However, lens distortions can change by zooming, focusing and temperature variations. Moreover, when only the video sequence is available preliminary calibration is often not an option. This paper addresses all fore-mentioned problems by sequentially recovering lens distortions together with structure and motion from video sequences without tedious pre-calibrations and allowing lens distortions to change over time. The devised algorithms are fairly simple as they only use linear least squares techniques. The unprocessed video sequence forms the only input and no severe restrictions are placed on viewed scene geometry. Therefore, the accurate recovery of structure and motion is fully automated and widely applicable. The experiments demonstrate the necessity of modeling lens distortions to achieve high accuracy in recovered structure and motion.

DLOW: Domain Flow and Applications

Abstract: In this work, we present a domain flow generation (DLOW) model to bridge two different domains by generating a continuous sequence of intermediate domains flowing from one domain to the other. The benefits of our DLOW model are twofold. First, it is able to transfer source images into a domain flow, which consists of images with smoothly changing distributions from the source to the target domain. The domain flow bridges the gap between source and target domains, thus easing the domain adaptation task. Second, when multiple target domains are provided for training, our DLOW model is also able to generate new styles of images that are unseen in the training data. The new images are shown to be able to mimic different artists to produce a natural blend of multiple art styles. Furthermore, for the semantic segmentation in the adverse weather condition, we take advantage of our DLOW model to generate images with gradually changing fog density, which can be readily used for boosting the segmentation performance when combined with a curriculum learning strategy. We demonstrate the effectiveness of our model on benchmark datasets for different applications, including cross-domain semantic segmentation, style generalization, and foggy scene understanding. Our implementation is available at https://github.com/ETHRuiGong/DLOW .

Deep Residual Learning for Image Recognition

Abstract: 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.

Very Deep Convolutional Networks for Large-Scale Image Recognition

Abstract: 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

Abstract: 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.

Deep Residual Learning for Image Recognition

Abstract: 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.

Going deeper with convolutions

Abstract: 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.