October 8, 2016

ECCV Brings Together the Brightest Minds in Computer Vision

By: Kelly Berschauer

The 14th European Conference on Computer Vision (ECCV) is happening this week in Amsterdam. As one of the top conferences in computer vision, Facebook researchers are on hand to learn from their peers and share their ideas and research though papers, posters as well as bring special interest groups together at focused workshops and tutorials.

Facebook AI Research (FAIR) Scientists Pedro O. Pinheiro, Tsung-Yi Lin, Ronan Collobert, and Piotr Dollár will be presenting their paper Learning to Refine Object Segments, which was highlighted in a blog earlier this year. The paper proposes to augment feedforward nets for object segmentation with a novel top-down refinement approach. Through this simple, fast, effective approach, the paper demonstrates how the resulting bottom-up/top-down architecture is capable of efficiently generating high-fidelity object masks.

In addition to presenting their paper, the team has made the code for DeepMask+SharpMask as well as MultiPathNet, and the demos related to them, open and accessible to all, with the hope that they’ll help rapidly advance the field of machine vision.

Of particular note this year is the 2nd ImageNet and COCO Visual Recognition Challenges Joint Workshop being run by Facebook Artificial Intelligence Researchers Ross Girshick and Piotr Dollár along with Tsung-Yi Lin and Yin Cui of Cornell and Genevieve Patterson of Brown and Matteo Ruggero Ronchi of Caltech. “For me, the workshops are the most interesting events of the conference, and the place where I see the most bleeding edge work, especially those that have a challenge associated with them.” Said Ross Girshick. “These competitive challenges are developed in broad collaboration with researchers from academia and industry and really help shape the direction of future work in visual recognition.”

The purpose of the workshop is to present the methods and results of the ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2016 and Common Objects in Context (COCO) 2016 Detection Challenge. Challenge participants with the most successful and innovative entries have been invited to share their work with the broader community. A new addition this year is the COCO 2016 Keypoint Challenge which requires localization of person keypoints in challenging, uncontrolled conditions. The Keypoint Challenge uses a fairly under-explored setting, involving simultaneously detecting people and localizing their keypoints, so the results should be exciting to see.

Other Facebook participation at ECCV 2016 includes:

Facebook AI Research Scientist, Laurens van der Maaten is delivering an Invited talk, Learning to Solve Vision without Annotating Millions of Images, during the Workshop on Web-scale Vision and Social Media.

Poster session presentations:

Learning to Refine Object Segments, Pedro Pinheiro, EPFL, Tsung-Yi Lin, Cornell, Ronan Collobert and Piotr Dollar of Facebook

Polysemous Codes, Matthijs Douze, Herve Jegou, and Florent Perronnin, Facebook

Learning Visual Features from Large Weakly Supervised Data, Armand Joulin, Laurens van der Maaten, Allan Jabri, and Nicolas Vasilache, Facebook AI Research

Revisiting Visual Question Answering Baselines, Allan Jabri, Armand Joulin, and Laurens van der Maaten, Facebook AI Research

Shuffle and Learn: Unsupervised Learning using Temporal Order Verification, Ishan Misra, CMU, Larry Zitnick, Facebook, and Martial Hebert, Carnegie Mellon University