Search Results for author: Mathieu Salzmann

Found 134 papers, 43 papers with code

On the Impact of Hard Adversarial Instances on Overfitting in Adversarial Training

no code implementations14 Dec 2021 Chen Liu, Zhichao Huang, Mathieu Salzmann, Tong Zhang, Sabine Süsstrunk

This lets us show that the decay in generalization performance of adversarial training is a result of the model's attempt to fit hard adversarial instances.

Adversarial Parametric Pose Prior

no code implementations8 Dec 2021 Andrey Davydov, Anastasia Remizova, Victor Constantin, Sina Honari, Mathieu Salzmann, Pascal Fua

The Skinned Multi-Person Linear (SMPL) model can represent a human body by mapping pose and shape parameters to body meshes.

3D Reconstruction

Dyadic Human Motion Prediction

no code implementations1 Dec 2021 Isinsu Katircioglu, Costa Georgantas, Mathieu Salzmann, Pascal Fua

To evaluate this, and because no existing motion prediction datasets depict two closely-interacting subjects, we introduce the LindyHop600K dance dataset.

Human motion prediction Motion Forecasting +1

Learning Transferable Adversarial Perturbations

1 code implementation NeurIPS 2021 Krishna Kanth Nakka, Mathieu Salzmann

While effective, deep neural networks (DNNs) are vulnerable to adversarial attacks.

What Stops Learning-based 3D Registration from Working in the Real World?

no code implementations19 Nov 2021 Zheng Dang, Lizhou Wang, Junning Qiu, Minglei Lu, Mathieu Salzmann

We summarise our findings into a set of guidelines and demonstrate their effectiveness by applying them to different baseline methods, DCP and IDAM.

Point Cloud Registration

Temporally-Consistent Surface Reconstruction using Metrically-Consistent Atlases

1 code implementation12 Nov 2021 Jan Bednarik, Noam Aigerman, Vladimir G. Kim, Siddhartha Chaudhuri, Shaifali Parashar, Mathieu Salzmann, Pascal Fua

The key to making these correspondences semantically meaningful is to guarantee that the metric tensors computed at corresponding points are as similar as possible.

Surface Reconstruction

Estimating Image Depth in the Comics Domain

no code implementations7 Oct 2021 Deblina Bhattacharjee, Martin Everaert, Mathieu Salzmann, Sabine Süsstrunk

Estimating the depth of comics images is challenging as such images a) are monocular; b) lack ground-truth depth annotations; c) differ across different artistic styles; d) are sparse and noisy.

Depth Estimation Translation +1

An Analysis of Super-Net Heuristics in Weight-Sharing NAS

no code implementations4 Oct 2021 Kaicheng Yu, René Ranftl, Mathieu Salzmann

Weight sharing promises to make neural architecture search (NAS) tractable even on commodity hardware.

Neural Architecture Search

Generating Smooth Pose Sequences for Diverse Human Motion Prediction

1 code implementation ICCV 2021 Wei Mao, Miaomiao Liu, Mathieu Salzmann

Recent progress in stochastic motion prediction, i. e., predicting multiple possible future human motions given a single past pose sequence, has led to producing truly diverse future motions and even providing control over the motion of some body parts.

Human motion prediction motion prediction

Multi-level Motion Attention for Human Motion Prediction

1 code implementation17 Jun 2021 Wei Mao, Miaomiao Liu, Mathieu Salzmann, Hongdong Li

Whether based on recurrent or feed-forward neural networks, existing learning based methods fail to model the observation that human motion tends to repeat itself, even for complex sports actions and cooking activities.

Graph Convolutional Network Human motion prediction +1

Attention-based Domain Adaptation for Single Stage Detectors

no code implementations14 Jun 2021 Vidit Vidit, Mathieu Salzmann

While domain adaptation has been used to improve the performance of object detectors when the training and test data follow different distributions, previous work has mostly focused on two-stage detectors.

Domain Adaptation

Distilling Image Classifiers in Object Detectors

1 code implementation NeurIPS 2021 Shuxuan Guo, Jose M. Alvarez, Mathieu Salzmann

Knowledge distillation constitutes a simple yet effective way to improve the performance of a compact student network by exploiting the knowledge of a more powerful teacher.

Knowledge Distillation Object Detection +1

DAAIN: Detection of Anomalous and Adversarial Input using Normalizing Flows

1 code implementation30 May 2021 Samuel von Baußnern, Johannes Otterbach, Adrian Loy, Mathieu Salzmann, Thomas Wollmann

We demonstrate the effectiveness of our approach using an ESPNet trained on the Cityscapes dataset as segmentation model, an affine Normalizing Flow as density estimator and use blue noise to ensure homogeneous sampling.

Semantic Segmentation

SegmentMeIfYouCan: A Benchmark for Anomaly Segmentation

2 code implementations30 Apr 2021 Robin Chan, Krzysztof Lis, Svenja Uhlemeyer, Hermann Blum, Sina Honari, Roland Siegwart, Pascal Fua, Mathieu Salzmann, Matthias Rottmann

State-of-the-art semantic or instance segmentation deep neural networks (DNNs) are usually trained on a closed set of semantic classes.

Instance Segmentation Semantic Segmentation

Temporally-Coherent Surface Reconstruction via Metric-Consistent Atlases

1 code implementation ICCV 2021 Jan Bednarik, Vladimir G. Kim, Siddhartha Chaudhuri, Shaifali Parashar, Mathieu Salzmann, Pascal Fua, Noam Aigerman

We propose a method for the unsupervised reconstruction of a temporally-coherent sequence of surfaces from a sequence of time-evolving point clouds, yielding dense, semantically meaningful correspondences between all keyframes.

Surface Reconstruction

Landmark Regularization: Ranking Guided Super-Net Training in Neural Architecture Search

1 code implementation CVPR 2021 Kaicheng Yu, Rene Ranftl, Mathieu Salzmann

Weight sharing has become a de facto standard in neural architecture search because it enables the search to be done on commodity hardware.

Neural Architecture Search

Robust Differentiable SVD

2 code implementations8 Apr 2021 Wei Wang, Zheng Dang, Yinlin Hu, Pascal Fua, Mathieu Salzmann

Eigendecomposition of symmetric matrices is at the heart of many computer vision algorithms.

Image Classification Style Transfer

Modeling Object Dissimilarity for Deep Saliency Prediction

no code implementations8 Apr 2021 Bahar Aydemir, Deblina Bhattacharjee, Seungryong Kim, Tong Zhang, Mathieu Salzmann, Sabine Süsstrunk

Saliency prediction has made great strides over the past two decades, with current techniques modeling low-level information, such as color, intensity and size contrasts, and high-level one, such as attention and gaze direction for entire objects.

Saliency Prediction

Progressive Correspondence Pruning by Consensus Learning

no code implementations ICCV 2021 Chen Zhao, Yixiao Ge, Feng Zhu, Rui Zhao, Hongsheng Li, Mathieu Salzmann

Correspondence selection aims to correctly select the consistent matches (inliers) from an initial set of putative correspondences.

Denoising Pose Estimation

Learning to Generate the Unknowns for Open-set Domain Adaptation

no code implementations1 Jan 2021 Mahsa Baktashmotlagh, Tianle Chen, Mathieu Salzmann

In this setting, existing techniques focus on the challenging task of isolating the unknown target samples, so as to avoid the negative transfer resulting from aligning the source feature distributions with the broader target one that encompasses the additional unknown classes.

Domain Adaptation

Temporally-Transferable Perturbations: Efficient, One-Shot Adversarial Attacks for Online Visual Object Trackers

no code implementations30 Dec 2020 Krishna Kanth Nakka, Mathieu Salzmann

While these methods were shown to be vulnerable to adversarial attacks, as most deep networks for visual recognition tasks, the existing attacks for VOT trackers all require perturbing the search region of every input frame to be effective, which comes at a non-negligible cost, considering that VOT is a real-time task.

Visual Object Tracking

Detecting Road Obstacles by Erasing Them

no code implementations25 Dec 2020 Krzysztof Lis, Sina Honari, Pascal Fua, Mathieu Salzmann

Vehicles can encounter a myriad of obstacles on the road, and it is impossible to record them all beforehand to train a detector.

Unsupervised Domain Adaptation with Temporal-Consistent Self-Training for 3D Hand-Object Joint Reconstruction

no code implementations21 Dec 2020 Mengshi Qi, Edoardo Remelli, Mathieu Salzmann, Pascal Fua

Deep learning-solutions for hand-object 3D pose and shape estimation are now very effective when an annotated dataset is available to train them to handle the scenarios and lighting conditions they will encounter at test time.

Unsupervised Domain Adaptation

Human Detection and Segmentation via Multi-view Consensus

1 code implementation ICCV 2021 Isinsu Katircioglu, Helge Rhodin, Jörg Spörri, Mathieu Salzmann, Pascal Fua

Self-supervised detection and segmentation of foreground objects aims for accuracy without annotated training data.

Human Detection

Long Term Motion Prediction Using Keyposes

1 code implementation8 Dec 2020 Sena Kiciroglu, Wei Wang, Mathieu Salzmann, Pascal Fua

Long term human motion prediction is essential in safety-critical applications such as human-robot interaction and autonomous driving.

Autonomous Driving Human motion prediction +2

Probabilistic Tracklet Scoring and Inpainting for Multiple Object Tracking

no code implementations CVPR 2021 Fatemeh Saleh, Sadegh Aliakbarian, Hamid Rezatofighi, Mathieu Salzmann, Stephen Gould

Despite the recent advances in multiple object tracking (MOT), achieved by joint detection and tracking, dealing with long occlusions remains a challenge.

Multiple Object Tracking

Counting People by Estimating People Flows

1 code implementation1 Dec 2020 Weizhe Liu, Mathieu Salzmann, Pascal Fua

Modern methods for counting people in crowded scenes rely on deep networks to estimate people densities in individual images.

Active Learning Crowd Counting +1

PCLs: Geometry-aware Neural Reconstruction of 3D Pose with Perspective Crop Layers

1 code implementation CVPR 2021 Frank Yu, Mathieu Salzmann, Pascal Fua, Helge Rhodin

Our conclusion is that it is important to utilize camera calibration information when available, for classical and deep-learning-based computer vision alike.

3D Reconstruction

A Closed-Form Solution to Local Non-Rigid Structure-from-Motion

no code implementations23 Nov 2020 Shaifali Parashar, Yuxuan Long, Mathieu Salzmann, Pascal Fua

A recent trend in Non-Rigid Structure-from-Motion (NRSfM) is to express local, differential constraints between pairs of images, from which the surface normal at any point can be obtained by solving a system of polynomial equations.

3D Registration for Self-Occluded Objects in Context

no code implementations23 Nov 2020 Zheng Dang, Fei Wang, Mathieu Salzmann

While much progress has been made on the task of 3D point cloud registration, there still exists no learning-based method able to estimate the 6D pose of an object observed by a 2. 5D sensor in a scene.

Instance Segmentation Point Cloud Registration +2

Self-supervised Segmentation via Background Inpainting

no code implementations11 Nov 2020 Isinsu Katircioglu, Helge Rhodin, Victor Constantin, Jörg Spörri, Mathieu Salzmann, Pascal Fua

While supervised object detection and segmentation methods achieve impressive accuracy, they generalize poorly to images whose appearance significantly differs from the data they have been trained on.

Human Detection Object Detection +1

Better Patch Stitching for Parametric Surface Reconstruction

1 code implementation14 Oct 2020 Zhantao Deng, Jan Bednařík, Mathieu Salzmann, Pascal Fua

We introduce an approach that explicitly encourages global consistency of the local mappings.

Surface Reconstruction

Motion Prediction Using Temporal Inception Module

1 code implementation6 Oct 2020 Tim Lebailly, Sena Kiciroglu, Mathieu Salzmann, Pascal Fua, Wei Wang

We argue that the diverse temporal scales are important as they allow us to look at the past frames with different receptive fields, which can lead to better predictions.

Autonomous Driving Human motion prediction +1

Robust RGB-based 6-DoF Pose Estimation without Real Pose Annotations

no code implementations19 Aug 2020 Zhigang Li, Yinlin Hu, Mathieu Salzmann, Xiangyang Ji

We achieve state of the art performance on LINEMOD, and OccludedLINEMOD in without real-pose setting, even outperforming methods that rely on real annotations during training on Occluded-LINEMOD.

Pose Estimation

GarNet++: Improving Fast and Accurate Static3D Cloth Draping by Curvature Loss

no code implementations20 Jul 2020 Erhan Gundogdu, Victor Constantin, Shaifali Parashar, Amrollah Seifoddini, Minh Dang, Mathieu Salzmann, Pascal Fua

We introduce a two-stream deep network model that produces a visually plausible draping of a template cloth on virtual 3D bodies by extracting features from both the body and garment shapes.

Volumetric Transformer Networks

no code implementations ECCV 2020 Seungryong Kim, Sabine Süsstrunk, Mathieu Salzmann

We design our VTN as an encoder-decoder network, with modules dedicated to letting the information flow across the feature channels, to account for the dependencies between the semantic parts.

Fine-Grained Image Recognition Image Retrieval

Towards Robust Fine-grained Recognition by Maximal Separation of Discriminative Features

no code implementations10 Jun 2020 Krishna Kanth Nakka, Mathieu Salzmann

In this paper, we identify the proximity of the latent representations of different classes in fine-grained recognition networks as a key factor to the success of adversarial attacks.

General Classification

Learning 3D-3D Correspondences for One-shot Partial-to-partial Registration

no code implementations8 Jun 2020 Zheng Dang, Fei Wang, Mathieu Salzmann

While 3D-3D registration is traditionally tacked by optimization-based methods, recent work has shown that learning-based techniques could achieve faster and more robust results.

Domain Adaptive Multibranch Networks

no code implementations ICLR 2020 Róger Bermúdez-Chacón, Mathieu Salzmann, Pascal Fua

We tackle unsupervised domain adaptation by accounting for the fact that different domains may need to be processed differently to arrive to a common feature representation effective for recognition.

Unsupervised Domain Adaptation

ArTIST: Autoregressive Trajectory Inpainting and Scoring for Tracking

no code implementations16 Apr 2020 Fatemeh Saleh, Sadegh Aliakbarian, Mathieu Salzmann, Stephen Gould

One of the core components in online multiple object tracking (MOT) frameworks is associating new detections with existing tracklets, typically done via a scoring function.

Human motion prediction motion prediction +1

Eigendecomposition-Free Training of Deep Networks for Linear Least-Square Problems

no code implementations15 Apr 2020 Zheng Dang, Kwang Moo Yi, Yinlin Hu, Fei Wang, Pascal Fua, Mathieu Salzmann

In this paper, we introduce an eigendecomposition-free approach to training a deep network whose loss depends on the eigenvector corresponding to a zero eigenvalue of a matrix predicted by the network.

Denoising Pose Estimation

How to Train Your Super-Net: An Analysis of Training Heuristics in Weight-Sharing NAS

no code implementations9 Mar 2020 Kaicheng Yu, Rene Ranftl, Mathieu Salzmann

Weight sharing promises to make neural architecture search (NAS) tractable even on commodity hardware.

Neural Architecture Search

Contextually Plausible and Diverse 3D Human Motion Prediction

no code implementations ICCV 2021 Sadegh Aliakbarian, Fatemeh Sadat Saleh, Lars Petersson, Stephen Gould, Mathieu Salzmann

We tackle the task of diverse 3D human motion prediction, that is, forecasting multiple plausible future 3D poses given a sequence of observed 3D poses.

Human motion prediction Image Captioning +1

Training Provably Robust Models by Polyhedral Envelope Regularization

1 code implementation10 Dec 2019 Chen Liu, Mathieu Salzmann, Sabine Süsstrunk

Training certifiable neural networks enables one to obtain models with robustness guarantees against adversarial attacks.

Indirect Local Attacks for Context-aware Semantic Segmentation Networks

1 code implementation ECCV 2020 Krishna Kanth Nakka, Mathieu Salzmann

Recently, deep networks have achieved impressive semantic segmentation performance, in particular thanks to their use of larger contextual information.

Semantic Segmentation

Using Depth for Pixel-Wise Detection of Adversarial Attacks in Crowd Counting

no code implementations26 Nov 2019 Weizhe Liu, Mathieu Salzmann, Pascal Fua

State-of-the-art methods for counting people in crowded scenes rely on deep networks to estimate crowd density.

Crowd Counting Density Estimation

Estimating People Flows to Better Count Them in Crowded Scenes

1 code implementation ECCV 2020 Weizhe Liu, Mathieu Salzmann, Pascal Fua

In this paper, we advocate estimating people flows across image locations between consecutive images and inferring the people densities from these flows instead of directly regressing.

Optical Flow Estimation

Shape Reconstruction by Learning Differentiable Surface Representations

1 code implementation CVPR 2020 Jan Bednarik, Shaifali Parashar, Erhan Gundogdu, Mathieu Salzmann, Pascal Fua

Generative models that produce point clouds have emerged as a powerful tool to represent 3D surfaces, and the best current ones rely on learning an ensemble of parametric representations.

Single-Stage 6D Object Pose Estimation

1 code implementation CVPR 2020 Yinlin Hu, Pascal Fua, Wei Wang, Mathieu Salzmann

Second, training the deep network relies on a surrogate loss that does not directly reflect the final 6D pose estimation task.

6D Pose Estimation 6D Pose Estimation using RGB

Learning Trajectory Dependencies for Human Motion Prediction

3 code implementations ICCV 2019 Wei Mao, Miaomiao Liu, Mathieu Salzmann, Hongdong Li

In this paper, we propose a simple feed-forward deep network for motion prediction, which takes into account both temporal smoothness and spatial dependencies among human body joints.

Graph Convolutional Network Human motion prediction +2

Learning Variations in Human Motion via Mix-and-Match Perturbation

no code implementations2 Aug 2019 Mohammad Sadegh Aliakbarian, Fatemeh Sadat Saleh, Mathieu Salzmann, Lars Petersson, Stephen Gould, Amirhossein Habibian

In this paper, we introduce an approach to stochastically combine the root of variations with previous pose information, which forces the model to take the noise into account.

Human motion prediction motion prediction

Self-supervised Training of Proposal-based Segmentation via Background Prediction

no code implementations18 Jul 2019 Isinsu Katircioglu, Helge Rhodin, Victor Constantin, Jörg Spörri, Mathieu Salzmann, Pascal Fua

While supervised object detection methods achieve impressive accuracy, they generalize poorly to images whose appearance significantly differs from the data they have been trained on.

Object Detection

Recurrent U-Net for Resource-Constrained Segmentation

no code implementations ICCV 2019 Wei Wang, Kaicheng Yu, Joachim Hugonot, Pascal Fua, Mathieu Salzmann

State-of-the-art segmentation methods rely on very deep networks that are not always easy to train without very large training datasets and tend to be relatively slow to run on standard GPUs.

Hand Segmentation

Detecting the Unexpected via Image Resynthesis

3 code implementations ICCV 2019 Krzysztof Lis, Krishna Nakka, Pascal Fua, Mathieu Salzmann

In this paper, we tackle the more realistic scenario where unexpected objects of unknown classes can appear at test time.

Resynthesis Semantic Segmentation

Neural Scene Decomposition for Multi-Person Motion Capture

1 code implementation CVPR 2019 Helge Rhodin, Victor Constantin, Isinsu Katircioglu, Mathieu Salzmann, Pascal Fua

To this end, we introduce a self-supervised approach to learning what we call a neural scene decomposition (NSD) that can be exploited for 3D pose estimation.

3D Pose Estimation Instance Segmentation +1

Overcoming Multi-Model Forgetting

no code implementations ICLR 2019 Yassine Benyahia, Kaicheng Yu, Kamil Bennani-Smires, Martin Jaggi, Anthony Davison, Mathieu Salzmann, Claudiu Musat

We identify a phenomenon, which we refer to as multi-model forgetting, that occurs when sequentially training multiple deep networks with partially-shared parameters; the performance of previously-trained models degrades as one optimizes a subsequent one, due to the overwriting of shared parameters.

Neural Architecture Search

Interpretable BoW Networks for Adversarial Example Detection

no code implementations8 Jan 2019 Krishna Kanth Nakka, Mathieu Salzmann

The reason behind the prediction for a new sample can then be interpreted by looking at the visual representation of the most highly activated codeword.

Segmentation-driven 6D Object Pose Estimation

5 code implementations CVPR 2019 Yinlin Hu, Joachim Hugonot, Pascal Fua, Mathieu Salzmann

The most recent trend in estimating the 6D pose of rigid objects has been to train deep networks to either directly regress the pose from the image or to predict the 2D locations of 3D keypoints, from which the pose can be obtained using a PnP algorithm.

6D Pose Estimation 6D Pose Estimation using RGB +1

Beyond One Glance: Gated Recurrent Architecture for Hand Segmentation

no code implementations27 Nov 2018 Wei Wang, Kaicheng Yu, Joachim Hugonot, Pascal Fua, Mathieu Salzmann

As evidenced by our results on standard hand segmentation benchmarks and on our own dataset, our approach outperforms these other, simpler recurrent segmentation techniques, as well as the state-of-the-art hand segmentation one.

Hand Segmentation Mixed Reality

Context-Aware Crowd Counting

1 code implementation CVPR 2019 Weizhe Liu, Mathieu Salzmann, Pascal Fua

State-of-the-art methods for counting people in crowded scenes rely on deep networks to estimate crowd density.

Crowd Counting

ExpandNets: Linear Over-parameterization to Train Compact Convolutional Networks

no code implementations NeurIPS 2020 Shuxuan Guo, Jose M. Alvarez, Mathieu Salzmann

As evidenced by our experiments, our approach outperforms both training the compact network from scratch and performing knowledge distillation from a teacher.

General Classification Knowledge Distillation +3

Tracing in 2D to Reduce the Annotation Effort for 3D Deep Delineation

1 code implementation26 Nov 2018 Mateusz Koziński, Agata Mosinska, Mathieu Salzmann, Pascal Fua

The difficulty of obtaining annotations to build training databases still slows down the adoption of recent deep learning approaches for biomedical image analysis.

Effective Use of Synthetic Data for Urban Scene Semantic Segmentation

no code implementations ECCV 2018 Fatemeh Sadat Saleh, Mohammad Sadegh Aliakbarian, Mathieu Salzmann, Lars Petersson, Jose M. Alvarez

Our approach builds on the observation that foreground and background classes are not affected in the same manner by the domain shift, and thus should be treated differently.

Domain Adaptation Semantic Segmentation

Learning Factorized Representations for Open-set Domain Adaptation

no code implementations ICLR 2019 Mahsa Baktashmotlagh, Masoud Faraki, Tom Drummond, Mathieu Salzmann

To this end, we rely on the intuition that the source and target samples depicting the known classes can be generated by a shared subspace, whereas the target samples from unknown classes come from a different, private subspace.

Domain Adaptation

Deep Attentional Structured Representation Learning for Visual Recognition

no code implementations14 May 2018 Krishna Kanth Nakka, Mathieu Salzmann

Structured representations, such as Bags of Words, VLAD and Fisher Vectors, have proven highly effective to tackle complex visual recognition tasks.

Representation Learning Scene Recognition

Geometry-aware Deep Network for Single-Image Novel View Synthesis

no code implementations CVPR 2018 Miaomiao Liu, Xuming He, Mathieu Salzmann

By contrast, in this paper, we propose to exploit the 3D geometry of the scene to synthesize a novel view.

Novel View Synthesis

Learning to Reconstruct Texture-less Deformable Surfaces from a Single View

1 code implementation23 Mar 2018 Jan Bednařík, Pascal Fua, Mathieu Salzmann

Recent years have seen the development of mature solutions for reconstructing deformable surfaces from a single image, provided that they are relatively well-textured.

3D Reconstruction

Eigendecomposition-free Training of Deep Networks with Zero Eigenvalue-based Losses

no code implementations ECCV 2018 Zheng Dang, Kwang Moo Yi, Yinlin Hu, Fei Wang, Pascal Fua, Mathieu Salzmann

Many classical Computer Vision problems, such as essential matrix computation and pose estimation from 3D to 2D correspondences, can be solved by finding the eigenvector corresponding to the smallest, or zero, eigenvalue of a matrix representing a linear system.

3D Pose Estimation

Statistically Motivated Second Order Pooling

1 code implementation23 Jan 2018 Kaicheng Yu, Mathieu Salzmann

Our approach is motivated by a statistical analysis of the network's activations, relying on operations that lead to a Gaussian-distributed final representation, as inherently used by first-order deep networks.

Residual Parameter Transfer for Deep Domain Adaptation

no code implementations CVPR 2018 Artem Rozantsev, Mathieu Salzmann, Pascal Fua

The goal of Deep Domain Adaptation is to make it possible to use Deep Nets trained in one domain where there is enough annotated training data in another where there is little or none.

Domain Adaptation

Learning to Find Good Correspondences

3 code implementations CVPR 2018 Kwang Moo Yi, Eduard Trulls, Yuki Ono, Vincent Lepetit, Mathieu Salzmann, Pascal Fua

We develop a deep architecture to learn to find good correspondences for wide-baseline stereo.

Compression-aware Training of Deep Networks

no code implementations NeurIPS 2017 Jose M. Alvarez, Mathieu Salzmann

In recent years, great progress has been made in a variety of application domains thanks to the development of increasingly deeper neural networks.

Soft Correspondences in Multimodal Scene Parsing

no code implementations28 Sep 2017 Sarah Taghavi Namin, Mohammad Najafi, Mathieu Salzmann, Lars Petersson

We propose to address this issue, by formulating multimodal semantic labeling as inference in a CRF and introducing latent nodes to explicitly model inconsistencies between two modalities.

Scene Parsing

Bringing Background into the Foreground: Making All Classes Equal in Weakly-supervised Video Semantic Segmentation

no code implementations ICCV 2017 Fatemeh Sadat Saleh, Mohammad Sadegh Aliakbarian, Mathieu Salzmann, Lars Petersson, Jose M. Alvarez

Our experiments demonstrate the benefits of our classifier heatmaps and of our two-stream architecture on challenging urban scene datasets and on the YouTube-Objects benchmark, where we obtain state-of-the-art results.

Autonomous Navigation Video Semantic Segmentation +1

Adaptive Low-Rank Kernel Subspace Clustering

1 code implementation17 Jul 2017 Pan Ji, Ian Reid, Ravi Garg, Hongdong Li, Mathieu Salzmann

In this paper, we present a kernel subspace clustering method that can handle non-linear models.

Image Clustering Motion Segmentation

Indoor Scene Parsing With Instance Segmentation, Semantic Labeling and Support Relationship Inference

no code implementations CVPR 2017 Wei Zhuo, Mathieu Salzmann, Xuming He, Miaomiao Liu

In particular, while some of them aim at segmenting the image into regions, such as object or surface instances, others aim at inferring the semantic labels of given regions, or their support relationships.

Instance Segmentation Scene Parsing +1

Imposing Hard Constraints on Deep Networks: Promises and Limitations

no code implementations7 Jun 2017 Pablo Márquez-Neila, Mathieu Salzmann, Pascal Fua

Imposing constraints on the output of a Deep Neural Net is one way to improve the quality of its predictions while loosening the requirements for labeled training data.

Incorporating Network Built-in Priors in Weakly-supervised Semantic Segmentation

no code implementations6 Jun 2017 Fatemeh Sadat Saleh, Mohammad Sadegh Aliakbarian, Mathieu Salzmann, Lars Petersson, Jose M. Alvarez, Stephen Gould

We then show how to obtain multi-class masks by the fusion of foreground/background ones with information extracted from a weakly-supervised localization network.

Object Recognition TAG +1

Encouraging LSTMs to Anticipate Actions Very Early

1 code implementation ICCV 2017 Mohammad Sadegh Aliakbarian, Fatemeh Sadat Saleh, Mathieu Salzmann, Basura Fernando, Lars Petersson, Lars Andersson

In contrast to the widely studied problem of recognizing an action given a complete sequence, action anticipation aims to identify the action from only partially available videos.

Action Anticipation Autonomous Navigation

Second-order Convolutional Neural Networks

no code implementations20 Mar 2017 Kaicheng Yu, Mathieu Salzmann

By performing linear combinations and element-wise nonlinear operations, these networks can be thought of as extracting solely first-order information from an input image.

Image Classification

Memory Efficient Max Flow for Multi-label Submodular MRFs

1 code implementation CVPR 2016 Thalaiyasingam Ajanthan, Richard Hartley, Mathieu Salzmann

Multi-label submodular Markov Random Fields (MRFs) have been shown to be solvable using max-flow based on an encoding of the labels proposed by Ishikawa, in which each variable $X_i$ is represented by $\ell$ nodes (where $\ell$ is the number of labels) arranged in a column.

Boundary-aware Instance Segmentation

no code implementations CVPR 2017 Zeeshan Hayder, Xuming He, Mathieu Salzmann

In this context, existing methods typically propose candidate objects, usually as bounding boxes, and directly predict a binary mask within each such proposal.

Instance Segmentation Object Proposal Generation +1

Efficient Linear Programming for Dense CRFs

no code implementations CVPR 2017 Thalaiyasingam Ajanthan, Alban Desmaison, Rudy Bunel, Mathieu Salzmann, Philip H. S. Torr, M. Pawan Kumar

To this end, we develop a proximal minimization framework, where the dual of each proximal problem is optimized via block coordinate descent.

Semantic Segmentation

Learning the Number of Neurons in Deep Networks

no code implementations NeurIPS 2016 Jose M. Alvarez, Mathieu Salzmann

In this paper, we introduce an approach to automatically determining the number of neurons in each layer of a deep network during learning.

Learning to Co-Generate Object Proposals With a Deep Structured Network

no code implementations CVPR 2016 Zeeshan Hayder, Xuming He, Mathieu Salzmann

In particular, we introduce a deep structured network that jointly predicts the objectness scores and the bounding box locations of multiple object candidates.

Object Detection

Semantic-Aware Depth Super-Resolution in Outdoor Scenes

no code implementations31 May 2016 Miaomiao Liu, Mathieu Salzmann, Xuming He

Despite much progress, state-of-the-art techniques suffer from two drawbacks: (i) they rely on the assumption that intensity edges coincide with depth discontinuities, which, unfortunately, is only true in controlled environments; and (ii) they typically exploit the availability of high-resolution training depth maps, which can often not be acquired in practice due to the sensors' limitations.

Super-Resolution

Dimensionality Reduction on SPD Manifolds: The Emergence of Geometry-Aware Methods

no code implementations20 May 2016 Mehrtash Harandi, Mathieu Salzmann, Richard Hartley

This lets us formulate dimensionality reduction as the problem of finding a projection that yields a low-dimensional manifold either with maximum discriminative power in the supervised scenario, or with maximum variance of the data in the unsupervised one.

Dimensionality Reduction

Beyond Sharing Weights for Deep Domain Adaptation

no code implementations21 Mar 2016 Artem Rozantsev, Mathieu Salzmann, Pascal Fua

To this end, we introduce a two-stream architecture, where one operates in the source domain and the other in the target domain.

Domain Adaptation Object Recognition

Robust Multi-body Feature Tracker: A Segmentation-free Approach

no code implementations CVPR 2016 Pan Ji, Hongdong Li, Mathieu Salzmann, Yiran Zhong

Feature tracking is a fundamental problem in computer vision, with applications in many computer vision tasks, such as visual SLAM and action recognition.

Action Recognition Motion Segmentation

Deformable 3D Fusion: From Partial Dynamic 3D Observations to Complete 4D Models

no code implementations ICCV 2015 Weipeng Xu, Mathieu Salzmann, Yongtian Wang, Yue Liu

Capturing the 3D motion of dynamic, non-rigid objects has attracted significant attention in computer vision.

Cutting Edge: Soft Correspondences in Multimodal Scene Parsing

no code implementations ICCV 2015 Sarah Taghavi Namin, Mohammad Najafi, Mathieu Salzmann, Lars Petersson

In this paper, we address the problem of data misalignment and label inconsistencies, e. g., due to moving objects, in semantic labeling, which violate the assumption of existing techniques.

Scene Parsing

Structural Kernel Learning for Large Scale Multiclass Object Co-Detection

no code implementations ICCV 2015 Zeeshan Hayder, Xuming He, Mathieu Salzmann

To exploit the correlations between objects, we build a fully-connected CRF on the candidates, which explicitly incorporates both geometric layout relations across object classes and similarity relations across multiple images.

Object Detection

Sample and Filter: Nonparametric Scene Parsing via Efficient Filtering

no code implementations CVPR 2016 Mohammad Najafi, Sarah Taghavi Namin, Mathieu Salzmann, Lars Petersson

By contrast, nonparametric approaches, which bypass any learning phase and directly transfer the labels from the training data to the query images, can readily exploit new labeled samples as they become available.

Scene Parsing Superpixels

Shape Interaction Matrix Revisited and Robustified: Efficient Subspace Clustering with Corrupted and Incomplete Data

1 code implementation ICCV 2015 Pan Ji, Mathieu Salzmann, Hongdong Li

The Shape Interaction Matrix (SIM) is one of the earliest approaches to performing subspace clustering (i. e., separating points drawn from a union of subspaces).

Face Clustering Motion Segmentation

When VLAD met Hilbert

no code implementations CVPR 2016 Mehrtash Harandi, Mathieu Salzmann, Fatih Porikli

Vectors of Locally Aggregated Descriptors (VLAD) have emerged as powerful image/video representations that compete with or even outperform state-of-the-art approaches on many challenging visual recognition tasks.

General Classification

Indoor Scene Structure Analysis for Single Image Depth Estimation

no code implementations CVPR 2015 Wei Zhuo, Mathieu Salzmann, Xuming He, Miaomiao Liu

We tackle the problem of single image depth estimation, which, without additional knowledge, suffers from many ambiguities.

Depth Estimation

Riemannian Coding and Dictionary Learning: Kernels to the Rescue

no code implementations CVPR 2015 Mehrtash Harandi, Mathieu Salzmann

While sparse coding on non-flat Riemannian manifolds has recently become increasingly popular, existing solutions either are dedicated to specific manifolds, or rely on optimization problems that are difficult to solve, especially when it comes to dictionary learning.

Dictionary Learning

A Framework for Shape Analysis via Hilbert Space Embedding

no code implementations13 Dec 2014 Sadeep Jayasumana, Mathieu Salzmann, Hongdong Li, Mehrtash Harandi

We propose a framework for 2D shape analysis using positive definite kernels defined on Kendall's shape manifold.

General Classification

Kernel Methods on the Riemannian Manifold of Symmetric Positive Definite Matrices

no code implementations CVPR 2013 Sadeep Jayasumana, Richard Hartley, Mathieu Salzmann, Hongdong Li, Mehrtash Harandi

To encode the geometry of the manifold in the mapping, we introduce a family of provably positive definite kernels on the Riemannian manifold of SPD matrices.

Motion Segmentation Pedestrian Detection +1

Optimizing Over Radial Kernels on Compact Manifolds

no code implementations CVPR 2014 Sadeep Jayasumana, Richard Hartley, Mathieu Salzmann, Hongdong Li, Mehrtash Harandi

We tackle the problem of optimizing over all possible positive definite radial kernels on Riemannian manifolds for classification.

General Classification

Kernel Methods on Riemannian Manifolds with Gaussian RBF Kernels

no code implementations30 Nov 2014 Sadeep Jayasumana, Richard Hartley, Mathieu Salzmann, Hongdong Li, Mehrtash Harandi

We then use the proposed framework to identify positive definite kernels on two specific manifolds commonly encountered in computer vision: the Riemannian manifold of symmetric positive definite matrices and the Grassmann manifold, i. e., the Riemannian manifold of linear subspaces of a Euclidean space.

Iteratively Reweighted Graph Cut for Multi-label MRFs with Non-convex Priors

no code implementations CVPR 2015 Thalaiyasingam Ajanthan, Richard Hartley, Mathieu Salzmann, Hongdong Li

While widely acknowledged as highly effective in computer vision, multi-label MRFs with non-convex priors are difficult to optimize.

Kernel Coding: General Formulation and Special Cases

no code implementations30 Aug 2014 Mehrtash Harandi, Mathieu Salzmann

In contrast, here, we study the problem of performing coding in a high-dimensional Hilbert space, where the classes are expected to be more easily separable.

Dictionary Learning

From Manifold to Manifold: Geometry-Aware Dimensionality Reduction for SPD Matrices

no code implementations4 Jul 2014 Mehrtash T. Harandi, Mathieu Salzmann, Richard Hartley

In particular, we search for a projection that yields a low-dimensional manifold with maximum discriminative power encoded via an affinity-weighted similarity measure based on metrics on the manifold.

Dimensionality Reduction

Expanding the Family of Grassmannian Kernels: An Embedding Perspective

no code implementations4 Jul 2014 Mehrtash T. Harandi, Mathieu Salzmann, Sadeep Jayasumana, Richard Hartley, Hongdong Li

Modeling videos and image-sets as linear subspaces has proven beneficial for many visual recognition tasks.

Domain Adaptation on the Statistical Manifold

no code implementations CVPR 2014 Mahsa Baktashmotlagh, Mehrtash T. Harandi, Brian C. Lovell, Mathieu Salzmann

Here, we propose to make better use of the structure of this manifold and rely on the distance on the manifold to compare the source and target distributions.

Object Recognition Unsupervised Domain Adaptation

Mirror Surface Reconstruction from a Single Image

no code implementations CVPR 2013 Miaomiao Liu, Richard Hartley, Mathieu Salzmann

In such conditions, our differential geometry analysis provides a theoretical proof that the shape of the mirror surface can be uniquely recovered if the pose of the reference target is known.

Surface Reconstruction

Continuous Inference in Graphical Models with Polynomial Energies

no code implementations CVPR 2013 Mathieu Salzmann

In this paper, we tackle the problem of performing inference in graphical models whose energy is a polynomial function of continuous variables.

Factorized Latent Spaces with Structured Sparsity

no code implementations NeurIPS 2010 Yangqing Jia, Mathieu Salzmann, Trevor Darrell

Recent approaches to multi-view learning have shown that factorizing the information into parts that are shared across all views and parts that are private to each view could effectively account for the dependencies and independencies between the different input modalities.

MULTI-VIEW LEARNING Pose Estimation

Cannot find the paper you are looking for? You can Submit a new open access paper.