Roey Mechrez

Senior researcher, PhD from the Technion-Israel, where he worked with Prof. Lihi Zelnik-Manor. Roey's work lies in the intersection of, computer vision, and deep learning. Specifically, he is interested in realistic image generation, manipulation and transformation, focusesing on tools, algorithms, and new paradigms for photo editing and synthesis. During his PhD research he has worked on Saliency Manipulation, Image Generation, Super-resolution and normal estimation, Photorealistic synthesis, and Template Matching.

His masters and bachelors were done at Tel Aviv University in Biomedical Engineering (cum laude).

               

News

I'm orginizing Deep learning Theory and Application for Computer Vision (winter school) at the Technion.
I'm organizing PIRM2018: Workshop and Challenge on Perceptual Image Restoration and Manipulation at ECCV 2018.

Research

I'm interested in computer vision, machine learning, optimization, image processing, and computational photography. Most of my papers deal with photorealistic image generation and manipulation in the low-mid level such as perceptual super resolution and in the high semantic level such as semantic style transfer.

Dynamic-Net: Tuning the Objective Without Re-training
Alon Shoshan, Roey Mechrez, Lihi Zelnik-Manor
ICCV, 2019
project page / arXiv / code / vision day 2018 talk

We present a first attempt at alleviating the need for re-training. Rather than fixing the network at training time, we train a ``Dynamic-Net'' that can be modified at inference time.

Adversarial Feedback Loop
Firas Shama, Roey Mechrez, Alon Shoshan, Lihi Zelnik-Manor
ICCV, 2019
project page / arXiv / code / vision day 2018 talk

We propose a novel method that makes an explicit use of the discriminator in test-time, in a feedback manner in order to improve the generator results.

Saliency Driven Image Manipulation (extended version)
Roey Mechrez, Eli Shechtman Lihi Zelnik-Manor
Machine Vision and Applications (special issue: IEEE WACV'18)
project page

The goal of this paper is to manipulating images in order to control the saliency of objects. We propose an approach that considers the internal color and saliency properties of the image.

Improving CNN Training using Disentanglement for Liver Lesion Classification in CT
Avi Ben-Cohen, Roey Mechrez, Noa Yedidia, Hayit Greenspan
To appear in EMBC 2019
arXiv

We suggest the synthesis of new data by mixing the class specified and unspecified representation of different factors in the training data.

Maintain Natural Image Statistics with the Contextual Loss
Roey Mechrez*, Itamar Talmi*, Firas Shama, Lihi Zelnik-Manor
ACCV, 2018
project page / GitHub / arXiv
Results: BSD100 / PIRM / DIV2K / set5 / set14 / urban100

We use the contextual loss for two image restoration problems: super-resolution and normal estimation and reveal its relation to the KL-divergence.

The 2018 PIRM Challenge on Perceptual Image Super-resolution
Yochai Blau*, Roey Mechrez*, Radu Timofte, Tomer Michaeli, Lihi Zelnik-Manor
ECCV workshop, 2018
project page / arXiv / all workshop papers at CVF

In contrast to previous SR challenges, our evaluation methodology jointly quantifies accuracy and perceptual quality

The Contextual Loss for Image Transformation with Non-Aligned Data
Roey Mechrez*, Itamar Talmi*, Lihi Zelnik-Manor
ECCV, 2018
project page / GitHub / arXiv / video / (Oral Acceptance rate 2.4%)
Resaults: faces / animals / cool politicians / single image animation / puppet control / male2female / female2male

We suggest a new loss function for image manipulation and generation. We show some very cool applications such as semantic style transfer and single image animations.

Saliency Driven Image Manipulation
Roey Mechrez, Eli Shechtman Lihi Zelnik-Manor
WACV, 2018
project page / GitHub / arXiv / video / Best paper (people's choice)

Manipulating images in order to control the saliency of objects is the goal of this paper. We propose an approach that considers the internal color and saliency properties of the image. It changes the saliency map via an optimization framework that relies on patch-based manipulation using only patches from within the same image to maintain its appearance characteristics.

Photorealistic Style Transfer with Screened Poisson Equation
Roey Mechrez, Eli Shechtman Lihi Zelnik-Manor
BMVC, 2018
project page / GitHub / arXiv / Results

Using the Screened Poisson Equation for photorealism style transfer

Template Matching with Deformable Diversity Similarity
Itamar Talmi*, Roey Mechrez*, Lihi Zelnik-Manor
CVPR, 2017
project page / GitHub / arXiv / (Spotlight Acceptance rate 8%)

Novel measure for template matching named Deformable Diversity Similarity -- based on the diversity of feature matches between a target image window and the template

Patch-Based Segmentation with Spatial Consistency: Application to MS Lesions in Brain MRI
Roey Mechrez, Jacob Goldberger Hayit Greenspan
International Journal of Biomedical Imaging, 2016
paper

MS Lesion Segmentation using a Multi-Channel Patch-Based Approach with Spatial Consistency
Roey Mechrez, Jacob Goldberger Hayit Greenspan
SPIE Medical Imaging. International Society for Optics and Photonics, 2015
Paper

Talks and Community service
computer_vision

Controlling the Latent Space at Inference Time -- Israel Vision Day 2018
Industry Talks at: Intel (Haifa), Huawei (London), Samsung (Tel-Aviv), GM (Herzelia)
Academy Talks at: HUJI and Weizmann
Photorealistic Image Synthesis and Mnipulation -- Israel Vision Day 2017
Template Matching with Deformable Diversity Similarity -- Israel Vision Day 2016
Reviewer: CVPR'18 (Outstanding Reviewer), ICCV'18, ECCV'18, PIRM'18 (area chair), CVPR'19, ICCV'19, BMVC'19, NTIRE'19 (program committee)

Teaching
computer_vision

EE046003 Deep learning Theory and Application for Computer Vision - Spring 2019

EE046746 Applications and Algorithms in Computer Vision - Spring 2018

EE046746 Applications and Algorithms in Computer Vision - Spring 2017