Iasonas Kokkinos

Stefano B. Blumberg, Daniele Raví, Mou-Cheng Xu et al.

Transposed convolution is crucial for generating high-resolution outputs, yet has received little attention compared to convolution layers. In this work we revisit transposed convolution and introduce a novel layer that allows us to place information in the image selectively and choose the `stroke breadth' at which the image is synthesized, whilst incurring a small additional parameter cost. For this we introduce three ideas: firstly, we regress offsets to the positions where the transpose convolution results are placed; secondly we broadcast the offset weight locations over a learnable neighborhood; and thirdly we use a compact parametrization to share weights and restrict offsets. We show that simply substituting upsampling operators with our novel layer produces substantial improvements across tasks as diverse as instance segmentation, object detection, semantic segmentation, generative image modeling, and 3D magnetic resonance image enhancement, while outperforming all existing variants of transposed convolutions. Our novel layer can be used as a drop-in replacement for 2D and 3D upsampling operators and the code will be publicly available.

Eleni Chiou, Eleftheria Panagiotaki, Iasonas Kokkinos

In this work we challenge the common approach of using a one-to-one mapping ('translation') between the source and target domains in unsupervised domain adaptation (UDA). Instead, we rely on stochastic translation to capture inherent translation ambiguities. This allows us to (i) train more accurate target networks by generating multiple outputs conditioned on the same source image, leveraging both accurate translation and data augmentation for appearance variability, (ii) impute robust pseudo-labels for the target data by averaging the predictions of a source network on multiple translated versions of a single target image and (iii) train and ensemble diverse networks in the target domain by modulating the degree of stochasticity in the translations. We report improvements over strong recent baselines, leading to state-of-the-art UDA results on two challenging semantic segmentation benchmarks. Our code is available at https://github.com/elchiou/Beyond-deterministic-translation-for-UDA.

Rıza Alp Güler, George Trigeorgis, Epameinondas Antonakos et al.

In this paper we propose to learn a mapping from image pixels into a dense template grid through a fully convolutional network. We formulate this task as a regression problem and train our network by leveraging upon manually annotated facial landmarks "in-the-wild". We use such landmarks to establish a dense correspondence field between a three-dimensional object template and the input image, which then serves as the ground-truth for training our regression system. We show that we can combine ideas from semantic segmentation with regression networks, yielding a highly-accurate "quantized regression" architecture. Our system, called DenseReg, allows us to estimate dense image-to-template correspondences in a fully convolutional manner. As such our network can provide useful correspondence information as a stand-alone system, while when used as an initialization for Statistical Deformable Models we obtain landmark localization results that largely outperform the current state-of-the-art on the challenging 300W benchmark. We thoroughly evaluate our method on a host of facial analysis tasks and also provide qualitative results for dense human body correspondence. We make our code available at http://alpguler.com/DenseReg.html along with supplementary materials.

Siddhartha Chandra, Iasonas Kokkinos

In this work we propose a structured prediction technique that combines the virtues of Gaussian Conditional Random Fields (G-CRF) with Deep Learning: (a) our structured prediction task has a unique global optimum that is obtained exactly from the solution of a linear system (b) the gradients of our model parameters are analytically computed using closed form expressions, in contrast to the memory-demanding contemporary deep structured prediction approaches that rely on back-propagation-through-time, (c) our pairwise terms do not have to be simple hand-crafted expressions, as in the line of works building on the DenseCRF, but can rather be `discovered' from data through deep architectures, and (d) out system can trained in an end-to-end manner. Building on standard tools from numerical analysis we develop very efficient algorithms for inference and learning, as well as a customized technique adapted to the semantic segmentation task. This efficiency allows us to explore more sophisticated architectures for structured prediction in deep learning: we introduce multi-resolution architectures to couple information across scales in a joint optimization framework, yielding systematic improvements. We demonstrate the utility of our approach on the challenging VOC PASCAL 2012 image segmentation benchmark, showing substantial improvements over strong baselines. We make all of our code and experiments available at {https://github.com/siddharthachandra/gcrf}

Eric-Tuan Lê, Antonis Kakolyris, Petros Koutras et al.

DensePose provides a pixel-accurate association of images with 3D mesh coordinates, but does not provide a 3D mesh, while Human Mesh Reconstruction (HMR) systems have high 2D reprojection error, as measured by DensePose localization metrics. In this work we introduce MeshPose to jointly tackle DensePose and HMR. For this we first introduce new losses that allow us to use weak DensePose supervision to accurately localize in 2D a subset of the mesh vertices ('VertexPose'). We then lift these vertices to 3D, yielding a low-poly body mesh ('MeshPose'). Our system is trained in an end-to-end manner and is the first HMR method to attain competitive DensePose accuracy, while also being lightweight and amenable to efficient inference, making it suitable for real-time AR applications.

Eleni Chiou, Francesco Giganti, Shonit Punwani et al.

Any novel medical imaging modality that differs from previous protocols e.g. in the number of imaging channels, introduces a new domain that is heterogeneous from previous ones. This common medical imaging scenario is rarely considered in the domain adaptation literature, which handles shifts across domains of the same dimensionality. In our work we rely on stochastic generative modeling to translate across two heterogeneous domains at pixel space and introduce two new loss functions that promote semantic consistency. Firstly, we introduce a semantic cycle-consistency loss in the source domain to ensure that the translation preserves the semantics. Secondly, we introduce a pseudo-labelling loss, where we translate target data to source, label them by a source-domain network, and use the generated pseudo-labels to supervise the target-domain network. Our results show that this allows us to extract systematically better representations for the target domain. In particular, we address the challenge of enhancing performance on VERDICT-MRI, an advanced diffusion-weighted imaging technique, by exploiting labeled mp-MRI data. When compared to several unsupervised domain adaptation approaches, our approach yields substantial improvements, that consistently carry over to the semi-supervised and supervised learning settings.

Filippos Kokkinos, Iasonas Kokkinos

We present To The Point (TTP), a method for reconstructing 3D objects from a single image using 2D to 3D correspondences learned from weak supervision. We recover a 3D shape from a 2D image by first regressing the 2D positions corresponding to the 3D template vertices and then jointly estimating a rigid camera transform and non-rigid template deformation that optimally explain the 2D positions through the 3D shape projection. By relying on 3D-2D correspondences we use a simple per-sample optimization problem to replace CNN-based regression of camera pose and non-rigid deformation and thereby obtain substantially more accurate 3D reconstructions. We treat this optimization as a differentiable layer and train the whole system in an end-to-end manner. We report systematic quantitative improvements on multiple categories and provide qualitative results comprising diverse shape, pose and texture prediction examples. Project website: https://fkokkinos.github.io/to_the_point/.

Filippos Kokkinos, Iasonas Kokkinos

Monocular 3D reconstruction of articulated object categories is challenging due to the lack of training data and the inherent ill-posedness of the problem. In this work we use video self-supervision, forcing the consistency of consecutive 3D reconstructions by a motion-based cycle loss. This largely improves both optimization-based and learning-based 3D mesh reconstruction. We further introduce an interpretable model of 3D template deformations that controls a 3D surface through the displacement of a small number of local, learnable handles. We formulate this operation as a structured layer relying on mesh-laplacian regularization and show that it can be trained in an end-to-end manner. We finally introduce a per-sample numerical optimisation approach that jointly optimises over mesh displacements and cameras within a video, boosting accuracy both for training and also as test time post-processing. While relying exclusively on a small set of videos collected per category for supervision, we obtain state-of-the-art reconstructions with diverse shapes, viewpoints and textures for multiple articulated object categories.

Eleni Chiou, Francesco Giganti, Shonit Punwani et al.

The need for training data can impede the adoption of novel imaging modalities for learning-based medical image analysis. Domain adaptation methods partially mitigate this problem by translating training data from a related source domain to a novel target domain, but typically assume that a one-to-one translation is possible. Our work addresses the challenge of adapting to a more informative target domain where multiple target samples can emerge from a single source sample. In particular we consider translating from mp-MRI to VERDICT, a richer MRI modality involving an optimized acquisition protocol for cancer characterization. We explicitly account for the inherent uncertainty of this mapping and exploit it to generate multiple outputs conditioned on a single input. Our results show that this allows us to extract systematically better image representations for the target domain, when used in tandem with both simple, CycleGAN-based baselines, as well as more powerful approaches that integrate discriminative segmentation losses and/or residual adapters. When compared to its deterministic counterparts, our approach yields substantial improvements across a broad range of dataset sizes, increasingly strong baselines, and evaluation measures.

Zbigniew Wojna, Krzysztof Maziarz, Łukasz Jocz et al.

We address six different classification tasks related to fine-grained building attributes: construction type, number of floors, pitch and geometry of the roof, facade material, and occupancy class. Tackling such a remote building analysis problem became possible only recently due to growing large-scale datasets of urban scenes. To this end, we introduce a new benchmarking dataset, consisting of 49426 images (top-view and street-view) of 9674 buildings. These photos are further assembled, together with the geometric metadata. The dataset showcases various real-world challenges, such as occlusions, blur, partially visible objects, and a broad spectrum of buildings. We propose a new projection pooling layer, creating a unified, top-view representation of the top-view and the side views in a high-dimensional space. It allows us to utilize the building and imagery metadata seamlessly. Introducing this layer improves classification accuracy -- compared to highly tuned baseline models -- indicating its suitability for building analysis.

Dominik Kulon, Riza Alp Güler, Iasonas Kokkinos et al.

We introduce a simple and effective network architecture for monocular 3D hand pose estimation consisting of an image encoder followed by a mesh convolutional decoder that is trained through a direct 3D hand mesh reconstruction loss. We train our network by gathering a large-scale dataset of hand action in YouTube videos and use it as a source of weak supervision. Our weakly-supervised mesh convolutions-based system largely outperforms state-of-the-art methods, even halving the errors on the in the wild benchmark. The dataset and additional resources are available at https://arielai.com/mesh_hands.

Eric-Tuan Le, Iasonas Kokkinos, Niloy J. Mitra

In this work we introduce Lean Point Networks (LPNs) to train deeper and more accurate point processing networks by relying on three novel point processing blocks that improve memory consumption, inference time, and accuracy: a convolution-type block for point sets that blends neighborhood information in a memory-efficient manner; a crosslink block that efficiently shares information across low- and high-resolution processing branches; and a multiresolution point cloud processing block for faster diffusion of information. By combining these blocks, we design wider and deeper point-based architectures. We report systematic accuracy and memory consumption improvements on multiple publicly available segmentation tasks by using our generic modules as drop-in replacements for the blocks of multiple architectures (PointNet++, DGCNN, SpiderNet, PointCNN).

Petru Manescu, Lydia Neary- Zajiczek, Michael J. Shaw et al.

Fast accurate diagnosis of malaria is still a global health challenge for which automated digital-pathology approaches could provide scalable solutions amenable to be deployed in low-to-middle income countries. Here we address the problem of Extended Depth-of-Field (EDoF) in thick blood film microscopy for rapid automated malaria diagnosis. High magnification oil-objectives (100x) with large numerical aperture are usually preferred to resolve the fine structural details that help separate true parasites from distractors. However, such objectives have a very limited depth-of-field requiring the acquisition of a series of images at different focal planes per field of view (FOV). Current EDoF techniques based on multi-scale decompositions are time consuming and therefore not suited for high-throughput analysis of specimens. To overcome this challenge, we developed a new deep learning method based on Convolutional Neural Networks (EDoF-CNN) that is able to rapidly perform the extended depth-of-field while also enhancing the spatial resolution of the resulting fused image. We evaluated our approach using simulated low-resolution z-stacks from Giemsa-stained thick blood smears from patients presenting with Plasmodium falciparum malaria. The EDoF-CNN allows speed-up of our digital-pathology acquisition platform and significantly improves the quality of the EDoF compared to the traditional multi-scaled approaches when applied to lower resolution stacks corresponding to acquisitions with fewer focal planes, large camera pixel binning or lower magnification objectives (larger FOV). We use the parasite detection accuracy of a deep learning model on the EDoFs as a concrete, task-specific measure of performance of this approach.

Natalia Neverova, James Thewlis, Rıza Alp Güler et al.

DensePose supersedes traditional landmark detectors by densely mapping image pixels to body surface coordinates. This power, however, comes at a greatly increased annotation time, as supervising the model requires to manually label hundreds of points per pose instance. In this work, we thus seek methods to significantly slim down the DensePose annotations, proposing more efficient data collection strategies. In particular, we demonstrate that if annotations are collected in video frames, their efficacy can be multiplied for free by using motion cues. To explore this idea, we introduce DensePose-Track, a dataset of videos where selected frames are annotated in the traditional DensePose manner. Then, building on geometric properties of the DensePose mapping, we use the video dynamic to propagate ground-truth annotations in time as well as to learn from Siamese equivariance constraints. Having performed exhaustive empirical evaluation of various data annotation and learning strategies, we demonstrate that doing so can deliver significantly improved pose estimation results over strong baselines. However, despite what is suggested by some recent works, we show that merely synthesizing motion patterns by applying geometric transformations to isolated frames is significantly less effective, and that motion cues help much more when they are extracted from videos.

Mihir Sahasrabudhe, Zhixin Shu, Edward Bartrum et al.

In this work we introduce Lifting Autoencoders, a generative 3D surface-based model of object categories. We bring together ideas from non-rigid structure from motion, image formation, and morphable models to learn a controllable, geometric model of 3D categories in an entirely unsupervised manner from an unstructured set of images. We exploit the 3D geometric nature of our model and use normal information to disentangle appearance into illumination, shading and albedo. We further use weak supervision to disentangle the non-rigid shape variability of human faces into identity and expression. We combine the 3D representation with a differentiable renderer to generate RGB images and append an adversarially trained refinement network to obtain sharp, photorealistic image reconstruction results. The learned generative model can be controlled in terms of interpretable geometry and appearance factors, allowing us to perform photorealistic image manipulation of identity, expression, 3D pose, and illumination properties.

Kevis-Kokitsi Maninis, Ilija Radosavovic, Iasonas Kokkinos

In this work we address task interference in universal networks by considering that a network is trained on multiple tasks, but performs one task at a time, an approach we refer to as "single-tasking multiple tasks". The network thus modifies its behaviour through task-dependent feature adaptation, or task attention. This gives the network the ability to accentuate the features that are adapted to a task, while shunning irrelevant ones. We further reduce task interference by forcing the task gradients to be statistically indistinguishable through adversarial training, ensuring that the common backbone architecture serving all tasks is not dominated by any of the task-specific gradients. Results in three multi-task dense labelling problems consistently show: (i) a large reduction in the number of parameters while preserving, or even improving performance and (ii) a smooth trade-off between computation and multi-task accuracy. We provide our system's code and pre-trained models at http://vision.ee.ethz.ch/~kmaninis/astmt/.

Maxim Berman, Hervé Jégou, Andrea Vedaldi et al.

MultiGrain is a network architecture producing compact vector representations that are suited both for image classification and particular object retrieval. It builds on a standard classification trunk. The top of the network produces an embedding containing coarse and fine-grained information, so that images can be recognized based on the object class, particular object, or if they are distorted copies. Our joint training is simple: we minimize a cross-entropy loss for classification and a ranking loss that determines if two images are identical up to data augmentation, with no need for additional labels. A key component of MultiGrain is a pooling layer that takes advantage of high-resolution images with a network trained at a lower resolution. When fed to a linear classifier, the learned embeddings provide state-of-the-art classification accuracy. For instance, we obtain 79.4% top-1 accuracy with a ResNet-50 learned on Imagenet, which is a +1.8% absolute improvement over the AutoAugment method. When compared with the cosine similarity, the same embeddings perform on par with the state-of-the-art for image retrieval at moderate resolutions.

Natalia Neverova, Riza Alp Guler, Iasonas Kokkinos

In this work we integrate ideas from surface-based modeling with neural synthesis: we propose a combination of surface-based pose estimation and deep generative models that allows us to perform accurate pose transfer, i.e. synthesize a new image of a person based on a single image of that person and the image of a pose donor. We use a dense pose estimation system that maps pixels from both images to a common surface-based coordinate system, allowing the two images to be brought in correspondence with each other. We inpaint and refine the source image intensities in the surface coordinate system, prior to warping them onto the target pose. These predictions are fused with those of a convolutional predictive module through a neural synthesis module allowing for training the whole pipeline jointly end-to-end, optimizing a combination of adversarial and perceptual losses. We show that dense pose estimation is a substantially more powerful conditioning input than landmark-, or mask-based alternatives, and report systematic improvements over state of the art generators on DeepFashion and MVC datasets.

Stefano B. Blumberg, Ryutaro Tanno, Iasonas Kokkinos et al.

In this paper we address the memory demands that come with the processing of 3-dimensional, high-resolution, multi-channeled medical images in deep learning. We exploit memory-efficient backpropagation techniques, to reduce the memory complexity of network training from being linear in the network's depth, to being roughly constant $ - $ permitting us to elongate deep architectures with negligible memory increase. We evaluate our methodology in the paradigm of Image Quality Transfer, whilst noting its potential application to various tasks that use deep learning. We study the impact of depth on accuracy and show that deeper models have more predictive power, which may exploit larger training sets. We obtain substantially better results than the previous state-of-the-art model with a slight memory increase, reducing the root-mean-squared-error by $ 13\% $. Our code is publicly available.

Siddhartha Chandra, Camille Couprie, Iasonas Kokkinos

In this work we introduce a time- and memory-efficient method for structured prediction that couples neuron decisions across both space at time. We show that we are able to perform exact and efficient inference on a densely connected spatio-temporal graph by capitalizing on recent advances on deep Gaussian Conditional Random Fields (GCRFs). Our method, called VideoGCRF is (a) efficient, (b) has a unique global minimum, and (c) can be trained end-to-end alongside contemporary deep networks for video understanding. We experiment with multiple connectivity patterns in the temporal domain, and present empirical improvements over strong baselines on the tasks of both semantic and instance segmentation of videos.

Zhixin Shu, Mihir Sahasrabudhe, Alp Guler et al.

In this work we introduce Deforming Autoencoders, a generative model for images that disentangles shape from appearance in an unsupervised manner. As in the deformable template paradigm, shape is represented as a deformation between a canonical coordinate system (`template') and an observed image, while appearance is modeled in `canonical', template, coordinates, thus discarding variability due to deformations. We introduce novel techniques that allow this approach to be deployed in the setting of autoencoders and show that this method can be used for unsupervised group-wise image alignment. We show experiments with expression morphing in humans, hands, and digits, face manipulation, such as shape and appearance interpolation, as well as unsupervised landmark localization. A more powerful form of unsupervised disentangling becomes possible in template coordinates, allowing us to successfully decompose face images into shading and albedo, and further manipulate face images.

Riza Alp Guler, Yuxiang Zhou, George Trigeorgis et al.

In this work we use deep learning to establish dense correspondences between a 3D object model and an image "in the wild". We introduce "DenseReg", a fully-convolutional neural network (F-CNN) that densely regresses at every foreground pixel a pair of U-V template coordinates in a single feedforward pass. To train DenseReg we construct a supervision signal by combining 3D deformable model fitting and 2D landmark annotations. We define the regression task in terms of the intrinsic, U-V coordinates of a 3D deformable model that is brought into correspondence with image instances at training time. A host of other object-related tasks (e.g. part segmentation, landmark localization) are shown to be by-products of this task, and to largely improve thanks to its introduction. We obtain highly-accurate regression results by combining ideas from semantic segmentation with regression networks, yielding a 'quantized regression' architecture that first obtains a quantized estimate of position through classification, and refines it through regression of the residual. We show that such networks can boost the performance of existing state-of-the-art systems for pose estimation. Firstly, we show that our system can serve as an initialization for Statistical Deformable Models, as well as an element of cascaded architectures that jointly localize landmarks and estimate dense correspondences. We also show that the obtained dense correspondence can act as a source of 'privileged information' that complements and extends the pure landmark-level annotations, accelerating and improving the training of pose estimation networks. We report state-of-the-art performance on the challenging 300W benchmark for facial landmark localization and on the MPII and LSP datasets for human pose estimation.

Rıza Alp Güler, Natalia Neverova, Iasonas Kokkinos

In this work, we establish dense correspondences between RGB image and a surface-based representation of the human body, a task we refer to as dense human pose estimation. We first gather dense correspondences for 50K persons appearing in the COCO dataset by introducing an efficient annotation pipeline. We then use our dataset to train CNN-based systems that deliver dense correspondence 'in the wild', namely in the presence of background, occlusions and scale variations. We improve our training set's effectiveness by training an 'inpainting' network that can fill in missing groundtruth values and report clear improvements with respect to the best results that would be achievable in the past. We experiment with fully-convolutional networks and region-based models and observe a superiority of the latter; we further improve accuracy through cascading, obtaining a system that delivers highly0accurate results in real time. Supplementary materials and videos are provided on the project page http://densepose.org

Neil Zeghidour, Nicolas Usunier, Iasonas Kokkinos et al.

We train a bank of complex filters that operates on the raw waveform and is fed into a convolutional neural network for end-to-end phone recognition. These time-domain filterbanks (TD-filterbanks) are initialized as an approximation of mel-filterbanks, and then fine-tuned jointly with the remaining convolutional architecture. We perform phone recognition experiments on TIMIT and show that for several architectures, models trained on TD-filterbanks consistently outperform their counterparts trained on comparable mel-filterbanks. We get our best performance by learning all front-end steps, from pre-emphasis up to averaging. Finally, we observe that the filters at convergence have an asymmetric impulse response, and that some of them remain almost analytic.

Adam W. Harley, Konstantinos G. Derpanis, Iasonas Kokkinos

We introduce an approach to integrate segmentation information within a convolutional neural network (CNN). This counter-acts the tendency of CNNs to smooth information across regions and increases their spatial precision. To obtain segmentation information, we set up a CNN to provide an embedding space where region co-membership can be estimated based on Euclidean distance. We use these embeddings to compute a local attention mask relative to every neuron position. We incorporate such masks in CNNs and replace the convolution operation with a "segmentation-aware" variant that allows a neuron to selectively attend to inputs coming from its own region. We call the resulting network a segmentation-aware CNN because it adapts its filters at each image point according to local segmentation cues. We demonstrate the merit of our method on two widely different dense prediction tasks, that involve classification (semantic segmentation) and regression (optical flow). Our results show that in semantic segmentation we can match the performance of DenseCRFs while being faster and simpler, and in optical flow we obtain clearly sharper responses than networks that do not use local attention masks. In both cases, segmentation-aware convolution yields systematic improvements over strong baselines. Source code for this work is available online at http://cs.cmu.edu/~aharley/segaware.

Natalia Neverova, Iasonas Kokkinos

Despite the large improvements in performance attained by using deep learning in computer vision, one can often further improve results with some additional post-processing that exploits the geometric nature of the underlying task. This commonly involves displacing the posterior distribution of a CNN in a way that makes it more appropriate for the task at hand, e.g. better aligned with local image features, or more compact. In this work we integrate this geometric post-processing within a deep architecture, introducing a differentiable and probabilistically sound counterpart to the common geometric voting technique used for evidence accumulation in vision. We refer to the resulting neural models as Mass Displacement Networks (MDNs), and apply them to human pose estimation in two distinct setups: (a) landmark localization, where we collapse a distribution to a point, allowing for precise localization of body keypoints and (b) communication across body parts, where we transfer evidence from one part to the other, allowing for a globally consistent pose estimate. We evaluate on large-scale pose estimation benchmarks, such as MPII Human Pose and COCO datasets, and report systematic improvements when compared to strong baselines.

Siddhartha Chandra, Iasonas Kokkinos

In this work we introduce a fully-connected graph structure in the Deep Gaussian Conditional Random Field (G-CRF) model. For this we express the pairwise interactions between pixels as the inner-products of low-dimensional embeddings, delivered by a new subnetwork of a deep architecture. We efficiently minimize the resulting energy by solving the resulting low-rank linear system with conjugate gradients, and derive an analytic expression for the gradient of our embeddings which allows us to train them end-to-end with backpropagation. We demonstrate the merit of our approach by achieving state of the art results on three challenging Computer Vision benchmarks, namely semantic segmentation, human parts segmentation, and saliency estimation. Our implementation is fully GPU based, built on top of the Caffe library, and will be made publicly available.

Iasonas Kokkinos

In this work we introduce a convolutional neural network (CNN) that jointly handles low-, mid-, and high-level vision tasks in a unified architecture that is trained end-to-end. Such a universal network can act like a `swiss knife' for vision tasks; we call this architecture an UberNet to indicate its overarching nature. We address two main technical challenges that emerge when broadening up the range of tasks handled by a single CNN: (i) training a deep architecture while relying on diverse training sets and (ii) training many (potentially unlimited) tasks with a limited memory budget. Properly addressing these two problems allows us to train accurate predictors for a host of tasks, without compromising accuracy. Through these advances we train in an end-to-end manner a CNN that simultaneously addresses (a) boundary detection (b) normal estimation (c) saliency estimation (d) semantic segmentation (e) human part segmentation (f) semantic boundary detection, (g) region proposal generation and object detection. We obtain competitive performance while jointly addressing all of these tasks in 0.7 seconds per frame on a single GPU. A demonstration of this system can be found at http://cvn.ecp.fr/ubernet/.

Mahsa Shakeri, Enzo Ferrante, Stavros Tsogkas et al.

We propose a modular and scalable framework for dense coregistration and cosegmentation with two key characteristics: first, we substitute ground truth data with the semantic map output of a classifier; second, we combine this output with population deformable registration to improve both alignment and segmentation. Our approach deforms all volumes towards consensus, taking into account image similarities and label consistency. Our pipeline can incorporate any classifier and similarity metric. Results on two datasets, containing annotations of challenging brain structures, demonstrate the potential of our method.

Liang-Chieh Chen, George Papandreou, Iasonas Kokkinos et al.

In this work we address the task of semantic image segmentation with Deep Learning and make three main contributions that are experimentally shown to have substantial practical merit. First, we highlight convolution with upsampled filters, or 'atrous convolution', as a powerful tool in dense prediction tasks. Atrous convolution allows us to explicitly control the resolution at which feature responses are computed within Deep Convolutional Neural Networks. It also allows us to effectively enlarge the field of view of filters to incorporate larger context without increasing the number of parameters or the amount of computation. Second, we propose atrous spatial pyramid pooling (ASPP) to robustly segment objects at multiple scales. ASPP probes an incoming convolutional feature layer with filters at multiple sampling rates and effective fields-of-views, thus capturing objects as well as image context at multiple scales. Third, we improve the localization of object boundaries by combining methods from DCNNs and probabilistic graphical models. The commonly deployed combination of max-pooling and downsampling in DCNNs achieves invariance but has a toll on localization accuracy. We overcome this by combining the responses at the final DCNN layer with a fully connected Conditional Random Field (CRF), which is shown both qualitatively and quantitatively to improve localization performance. Our proposed "DeepLab" system sets the new state-of-art at the PASCAL VOC-2012 semantic image segmentation task, reaching 79.7% mIOU in the test set, and advances the results on three other datasets: PASCAL-Context, PASCAL-Person-Part, and Cityscapes. All of our code is made publicly available online.

Mahsa Shakeri, Stavros Tsogkas, Enzo Ferrante et al.

In this paper we propose a deep learning approach for segmenting sub-cortical structures of the human brain in Magnetic Resonance (MR) image data. We draw inspiration from a state-of-the-art Fully-Convolutional Neural Network (F-CNN) architecture for semantic segmentation of objects in natural images, and adapt it to our task. Unlike previous CNN-based methods that operate on image patches, our model is applied on a full blown 2D image, without any alignment or registration steps at testing time. We further improve segmentation results by interpreting the CNN output as potentials of a Markov Random Field (MRF), whose topology corresponds to a volumetric grid. Alpha-expansion is used to perform approximate inference imposing spatial volumetric homogeneity to the CNN priors. We compare the performance of the proposed pipeline with a similar system using Random Forest-based priors, as well as state-of-art segmentation algorithms, and show promising results on two different brain MRI datasets.

Iasonas Kokkinos

In this work we show that adapting Deep Convolutional Neural Network training to the task of boundary detection can result in substantial improvements over the current state-of-the-art in boundary detection. Our contributions consist firstly in combining a careful design of the loss for boundary detection training, a multi-resolution architecture and training with external data to improve the detection accuracy of the current state of the art. When measured on the standard Berkeley Segmentation Dataset, we improve theoptimal dataset scale F-measure from 0.780 to 0.808 - while human performance is at 0.803. We further improve performance to 0.813 by combining deep learning with grouping, integrating the Normalized Cuts technique within a deep network. We also examine the potential of our boundary detector in conjunction with the task of semantic segmentation and demonstrate clear improvements over state-of-the-art systems. Our detector is fully integrated in the popular Caffe framework and processes a 320x420 image in less than a second.

Adam W. Harley, Konstantinos G. Derpanis, Iasonas Kokkinos

This paper proposes a new deep convolutional neural network (DCNN) architecture that learns pixel embeddings, such that pairwise distances between the embeddings can be used to infer whether or not the pixels lie on the same region. That is, for any two pixels on the same object, the embeddings are trained to be similar; for any pair that straddles an object boundary, the embeddings are trained to be dissimilar. Experimental results show that when this embedding network is used in conjunction with a DCNN trained on semantic segmentation, there is a systematic improvement in per-pixel classification accuracy. Our contributions are integrated in the popular Caffe deep learning framework, and consist in straightforward modifications to convolution routines. As such, they can be exploited for any task involving convolution layers.

Mircea Cimpoi, Subhransu Maji, Iasonas Kokkinos et al.

Visual textures have played a key role in image understanding because they convey important semantics of images, and because texture representations that pool local image descriptors in an orderless manner have had a tremendous impact in diverse applications. In this paper we make several contributions to texture understanding. First, instead of focusing on texture instance and material category recognition, we propose a human-interpretable vocabulary of texture attributes to describe common texture patterns, complemented by a new describable texture dataset for benchmarking. Second, we look at the problem of recognizing materials and texture attributes in realistic imaging conditions, including when textures appear in clutter, developing corresponding benchmarks on top of the recently proposed OpenSurfaces dataset. Third, we revisit classic texture representations, including bag-of-visual-words and the Fisher vectors, in the context of deep learning and show that these have excellent efficiency and generalization properties if the convolutional layers of a deep model are used as filter banks. We obtain in this manner state-of-the-art performance in numerous datasets well beyond textures, an efficient method to apply deep features to image regions, as well as benefit in transferring features from one domain to another.

Liang-Chieh Chen, George Papandreou, Iasonas Kokkinos et al.

Deep Convolutional Neural Networks (DCNNs) have recently shown state of the art performance in high level vision tasks, such as image classification and object detection. This work brings together methods from DCNNs and probabilistic graphical models for addressing the task of pixel-level classification (also called "semantic image segmentation"). We show that responses at the final layer of DCNNs are not sufficiently localized for accurate object segmentation. This is due to the very invariance properties that make DCNNs good for high level tasks. We overcome this poor localization property of deep networks by combining the responses at the final DCNN layer with a fully connected Conditional Random Field (CRF). Qualitatively, our "DeepLab" system is able to localize segment boundaries at a level of accuracy which is beyond previous methods. Quantitatively, our method sets the new state-of-art at the PASCAL VOC-2012 semantic image segmentation task, reaching 71.6% IOU accuracy in the test set. We show how these results can be obtained efficiently: Careful network re-purposing and a novel application of the 'hole' algorithm from the wavelet community allow dense computation of neural net responses at 8 frames per second on a modern GPU.

Edgar Simo-Serra, Eduard Trulls, Luis Ferraz et al.

In this paper we propose a novel framework for learning local image descriptors in a discriminative manner. For this purpose we explore a siamese architecture of Deep Convolutional Neural Networks (CNN), with a Hinge embedding loss on the L2 distance between descriptors. Since a siamese architecture uses pairs rather than single image patches to train, there exist a large number of positive samples and an exponential number of negative samples. We propose to explore this space with a stochastic sampling of the training set, in combination with an aggressive mining strategy over both the positive and negative samples which we denote as "fracking". We perform a thorough evaluation of the architecture hyper-parameters, and demonstrate large performance gains compared to both standard CNN learning strategies, hand-crafted image descriptors like SIFT, and the state-of-the-art on learned descriptors: up to 2.5x vs SIFT and 1.5x vs the state-of-the-art in terms of the area under the curve (AUC) of the Precision-Recall curve.

George Papandreou, Iasonas Kokkinos, Pierre-André Savalle

Deep Convolutional Neural Networks (DCNNs) commonly use generic `max-pooling' (MP) layers to extract deformation-invariant features, but we argue in favor of a more refined treatment. First, we introduce epitomic convolution as a building block alternative to the common convolution-MP cascade of DCNNs; while having identical complexity to MP, Epitomic Convolution allows for parameter sharing across different filters, resulting in faster convergence and better generalization. Second, we introduce a Multiple Instance Learning approach to explicitly accommodate global translation and scaling when training a DCNN exclusively with class labels. For this we rely on a `patchwork' data structure that efficiently lays out all image scales and positions as candidates to a DCNN. Factoring global and local deformations allows a DCNN to `focus its resources' on the treatment of non-rigid deformations and yields a substantial classification accuracy improvement. Third, further pursuing this idea, we develop an efficient DCNN sliding window object detector that employs explicit search over position, scale, and aspect ratio. We provide competitive image classification and localization results on the ImageNet dataset and object detection results on the Pascal VOC 2007 benchmark.

Mircea Cimpoi, Subhransu Maji, Iasonas Kokkinos et al.

Patterns and textures are defining characteristics of many natural objects: a shirt can be striped, the wings of a butterfly can be veined, and the skin of an animal can be scaly. Aiming at supporting this analytical dimension in image understanding, we address the challenging problem of describing textures with semantic attributes. We identify a rich vocabulary of forty-seven texture terms and use them to describe a large dataset of patterns collected in the wild.The resulting Describable Textures Dataset (DTD) is the basis to seek for the best texture representation for recognizing describable texture attributes in images. We port from object recognition to texture recognition the Improved Fisher Vector (IFV) and show that, surprisingly, it outperforms specialized texture descriptors not only on our problem, but also in established material recognition datasets. We also show that the describable attributes are excellent texture descriptors, transferring between datasets and tasks; in particular, combined with IFV, they significantly outperform the state-of-the-art by more than 8 percent on both FMD and KTHTIPS-2b benchmarks. We also demonstrate that they produce intuitive descriptions of materials and Internet images.