Intel AI Research at CVPR

June 16 – 20 in Long Beach, CA

Intel is a sponsor of the 2019 Conference on Computer Vision and Pattern Recognition (CVPR), with over 6,500 leading academics and researchers in computer vision, machine learning and artificial intelligence in attendance. Make the world your lab with Intel’s complete AI hardware portfolio for vision, backed by “write once, deploy anywhere” software and groundbreaking research.

CVPR 2019 Accepted Paper Presentations

Tuesday, June 18

PartNet: A Large-Scale Benchmark for Fine-Grained and Hierarchical Part-Level 3D Object Understanding

Time: 10:15 AM – 1:00 PM
Location: Poster #68 Paper ID: 177
Authors: Kaichun Mo – Stanford
Shilin Zhu – UCSD
Angel Chang – Simon Fraser
Li Yi – Stanford
Subarna Tripathi – Intel AI Lab
Leonidas J. Guibas – Stanford
Hao Su – UCSD
Abstract: We present PartNet: a consistent, large-scale dataset of 3D objects annotated with fine-grained, instance-level, and hierarchical 3D part information. We establish three benchmarking tasks for evaluating 3D part recognition using PartNet: fine-grained semantic segmentation, hierarchical semantic segmentation, and instance segmentation.

Tuesday, June 18

Zoom to Learn, Learn to Zoom

Time: 3:20 – 6:00 PM
Location: Poster #147, Paper ID: 1597
Authors: Xuaner Zhang – UC Berkeley
Qifeng Chen – HKUST
Ren Ng – UC Berkeley
Vladlen Koltun – Intel Labs
Abstract: We show that when applying machine learning to digital zoom for photography, it is beneficial to use real, RAW sensor data for training. The main barrier to using real sensor data for training is that ground truth high-resolution imagery is missing. We show how to obtain ground-truth data with optically zoomed images and contribute a dataset, SR-RAW, for real-world computational zoom.

Tuesday, June 18

What Do Single-view 3D Reconstruction Networks Learn?

Time: 3:20 – 6:00 PM
Location: Poster #93 Paper ID: 2029
Authors: Maxim Tatarchenko – University of Freiburg
Stephan Richter – Intel Labs
Rene Ranftl – Intel Labs
Zhuwen Li – Intel Labs
Vladlen Koltun – Intel Labs
Thomas Brox – University of Freiburg
Abstract: We show that encoder-decoder methods are statistically indistinguishable from baselines, thus indicating that the current state of the art in single-view object reconstruction does not actually perform reconstruction but image classification. We identify aspects of popular experimental procedures that elicit this behavior and discuss ways to improve the current state of research.

Tuesday, June 18

Events-to-Video: Bringing Modern Computer Vision to Event Cameras

Time: 3:20 – 6:00 PM
Location: Poster #157 Paper ID: 1595
Authors: Henri Rebecq – University of Zurich
Rene Ranftl – Intel Labs
Vladlen Koltun – Intel Labs
Davide Scaramuzza – University of Zurich
Abstract: Event cameras are novel sensors that report brightness changes in the form of asynchronous “events” instead of intensity frames. In this work, we apply existing, mature computer vision techniques to videos reconstructed from event data. We propose a novel recurrent network, and train it on a large amount of simulated event data. Our approach surpasses state-of-the-art reconstruction methods in terms of image quality.

Wednesday, June 19

Towards Accurate One-Stage Object Detection with AP-Loss

Time: 10:00 AM – 12:45 PM
Location: Poster #57, Paper ID: 2818
Authors: Kean Chen – SJTU
Jianguo Li – Intel Labs China
Weiyao Lin – SJTU
John See – MMU
Ji Wang – Tencent
Lingyu Duan – Peking University
Zhibo Chen – Tencent
Changwei He – Tencent
Junni Zou – SJTU
Abstract: This paper proposes a novel framework to replace the classification task in one-stage detectors with a ranking task, and adopting the Average-Precision loss (AP-loss) for the ranking problem.

Wednesday, June 19

Connecting the Dots: Learning Representations for Active Monocular Depth Estimation

Time: 3:20 – 6:00 PM
Location: Poster #90, Paper ID: 3965
Authors: Gernot Riegler – Intel Labs
Yiyi Liao – University of Tubingen
Simon Donné – University of Tubingen
Vladlen Koltun – Intel Labs
Andreas Geiger – University of Tubingen
Abstract: We propose a technique for depth estimation with a monocular structured-light camera, i.e., a calibrated stereo set-up with one camera and one laser projector. Instead of formulating the depth estimation via a correspondence search problem, we show that a simple convolutional architecture is sufficient for high-quality disparity estimates in this setting.

Wednesday, June 19

Acoustic Non-Line-of-Site Imaging

Time: 1:40 PM (ORAL); 3:20 – 6:00 PM (Poster)
Location: Poster # 132, Paper ID: 2059
Authors: David Lindell – Stanford
Gordon Wetzstein – Stanford
Vladlen Koltun – Intel Labs
Abstract: In this paper, we introduce acoustic NLOS imaging, which is orders of magnitude less expensive than most optical systems and captures hidden 3D geometry at longer ranges with shorter acquisition times compared to state-of-the-art optical methods.

Wednesday, June 19

Deeply-Supervised Knowledge Synergy

Time: 3:20 – 6:00 PM
Location: Poster #7, Paper ID: 3487
Authors: Dawei Sun – Intel Labs
Anbang Yao – Intel Labs
Aojun Zhou – Intel Labs
Hao Zhao – Intel Labs, Tsinghua University
Abstract: In this paper, we propose Deeply-supervised Knowledge Synergy (DKS), a new method aiming to train CNNs with improved generalization ability for image classification tasks without introducing extra computational cost during inference.

Thursday, June 20

CAM-Convs: Camera-Aware Multi-Scale Convolutions for Single-View Depth Prediction

Time: 3:20 – 6:00 PM
Location: Poster #87, Paper ID: 5655
Authors: José Fácil – University of Zaragoza
Huizhong Zhou – University of Freiburg
Benjamin Ummenhofer – Intel Labs
Thomas Brox – University of Freiburg
Javier Civera – University of Zaragoza
Luis Montesano – University of Zaragoza
Abstract: Single-view depth estimation suffers from the problem that a network trained on images from one camera does not generalize to images taken with a different camera model. In this work, we propose a new type of convolution that can take the camera parameters into account, thus allowing neural networks to learn calibration-aware patterns.

CVPR Workshops Poster Presentations

Sunday, June 16

PartNet: A Large-scale Benchmark for Fine-grained and Hierarchical Part-level 3D Object Understanding

Time: 10:05 AM – 11:00 AM and
3:05 PM – 4:00 PM
Location: 3D Scene Generation Workshop
Authors: Kaichun Mo – Stanford
Shilin Zhu – UCSD
Angel Chang – Simon Fraser
Li Yi – Stanford
Subarna Tripathi – Intel AI Lab
Leonidas J. Guibas – Stanford
Hao Su – UCSD
Abstract: We present PartNet: a consistent, large-scale dataset of 3D objects annotated with fine-grained, instance-level, and hierarchical 3D part information. We establish three benchmarking tasks for evaluating 3D part recognition using PartNet: fine-grained semantic segmentation, hierarchical semantic segmentation, and instance segmentation.

Sunday, June 16

To Believe or Not to Believe: Validating Explanation Fidelity for Dynamic Malware Analysis

Time: 11:30 AM – 1:00 PM
Location: Workshop on Explainable AI
Authors: Li Chen – Intel Labs
Carter Yagemann – Georgia Tech
Evan Downing – Georgia Tech
Abstract: In this work, via two case studies of dynamic malware classification, we extend the local interpretable model-agnostic explanation algorithm to explain image-based dynamic malware classification and examine its interpretation fidelity.

Sunday, June 16

Interpretable Machine Learning for Generating Semantically Meaningful Formative Feedback

Time: 11:30 AM – 1:00 PM
Location: Workshop on Explainable AI
Authors: Nese Alyuz – Intel
Tevfik Metin Sezgin – Koc University
Abstract: Research shows that children on the Autism Spectrum can be trained to recognize and express emotions if they are given supportive and constructive feedback. Today, formative feedback requires constant human supervision. In this work, we describe a system for automatic formative assessment integrated into an automatic emotion recognition setup.

Sunday, June 16

Using A Priori Knowledge to Improve Scene Understanding

Time: 3:00 – 3:50 PM
Location: Women in Computer Vision Workshop (Intel is a Silver Sponsor)
Authors: Brigit Schroeder – Intel AI Lab
Alex Alahi – EPFL
Abstract: Existing semantic segmentation algorithms treat images in isolation, but autonomous vehicles often revisit the same locations. We propose leveraging this a priori knowledge to improve semantic segmentation of images from sequential driving datasets.

Sunday, June 16

Using Scene Graph Context to Improve Image Generation

Time: 3:00 – 3:50 PM
Location: Women in Computer Vision Workshop (Intel is a Silver Sponsor)
Authors: Subarna Tripathi – Intel AI Lab
Anahita Bhiwandiwalla – Intel AI Lab
Alexei Bastidas – Intel AI Lab
Hanlin Tang – Intel AI Lab
Abstract: In this paper, we propose to harness scene graph context to improve image generation from scene graphs. We use both quantitative and qualitative studies to demonstrate that our proposed model outperforms the state-of-the-art on this challenging task.

Sunday, June 16

Compact Scene Graphs for Layout Composition and Patch Retrieval

Time: 3:25 – 4:25 PM
Location: CEFRL Workshop
Authors: Subarna Tripathi – Intel AI Lab
Sairam Sundaresan – Intel
Sharath Nittur Sridhar – Intel
Hanlin Tang – Intel AI Lab
Abstract: We propose two contributions to improve scene composition. First, we enhance the scene graph representation with heuristic-based relations, which add minimal storage overhead. Second, we use extreme points representation to supervise the learning of the scene composition network. These methods achieve significantly higher performance over existing methods.

Sunday, June 16

Channel Attention Networks

Time: 3:45 – 5:00 PM
Location: PBVS Workshop
Authors: Alexei Bastidas – Intel AI Lab
Hanlin Tang – Intel AI Lab
Abstract: In this work, we propose Channel Attention Networks (CAN), a deep learning model that uses soft attention on individual channels. We jointly train this model end-to-end on SpaceNet, a challenging multi-spectral semantic segmentation dataset.

Monday, June 17

PartNet: A Large-scale Benchmark for Fine-grained and Hierarchical Part-level 3D Object Understanding

Time: 10:15 AM – 11:00 PM and
3:30 PM – 4:00 PM
Location: Fine-Grained Visual Categorization workshop
Authors: Kaichun Mo – Stanford
Shilin Zhu – UCSD
Angel Chang – Simon Fraser
Li Yi – Stanford
Subarna Tripathi – Intel AI Lab
Leonidas J. Guibas – Stanford
Hao Su – UCSD
Abstract: We present PartNet: a consistent, large-scale dataset of 3D objects annotated with fine-grained, instance-level, and hierarchical 3D part information. We establish three benchmarking tasks for evaluating 3D part recognition using PartNet: fine-grained semantic segmentation, hierarchical semantic segmentation, and instance segmentation.

Monday, June 17

Context, Attention and Audio Feature Explorations for Audio Visual Scene-Aware Dialog

Time: 2:50 – 4:05 PM
Location: Visual Question Answering and Dialog Workshop
Authors: Shachi H. Kumar – Intel Labs
Eda Okur – Intel Labs
Saurav Sahay – Intel Labs
Juan Jose Alvarado Leanos – Intel Labs
Jonathan Huang – Intel Labs
Lama Nachman – Intel Labs
Abstract: This paper explores the role of ‘topics’ of dialog as the context of the conversation along with multimodal attention into an end-to-end audio-visual scene-aware dialog system architecture.

Monday, June 17

Uncertainty-Aware Audiovisual Activity Recognition using Deep Bayesian Variational Inference

Time: 3:10 – 4:00 PM
Location: Uncertainty and Robustness in Deep Visual Learning Workshop
Authors: Mahesh Subedar – Intel Labs
Ranganath Krishnan – Intel Labs
Paulo Lopez Meyer – Intel Labs
Omesh Tickoo – Intel Labs
Jonathan Huang – Intel Labs
Abstract: Deep neural networks (DNNs) provide state-of-the-art results for a multitude of applications, but the use of DNNs for multimodal audiovisual applications is still an unsolved problem. We apply Bayesian variational inference to DNNs for audiovisual activity recognition and quantify model uncertainty along with principled confidence. We propose a novel approach that combines deterministic and variational layers to estimate model uncertainty and principled confidence.

CVPR 2019 Expo: June 18 – 20, 2019

At the Intel booth, you’ll discover how Intel AI is breaking new ground in computer vision research. Join us for in-booth presentations, demonstrations, and the opportunity to connect with fellow researchers.

Booth Presentation Schedule

TIMESLOTS Tuesday, June 18 Wednesday, June 19 Thursday, June 20
10:15 –
10:45 AM
10:30 – 11:00 AM
Yash Akhuri (Intel AI Developer Community)
HadaNets: Flexible Quantization Strategies for Neural Networks
Philip Krejov (Intel)
Deep Learning for VR/AR: Body Tracking with Intel® RealSense™ Technology
Jahanzeb Ahmad (Intel): AI+: Combining AI and Other Critical Functions Using Intel® FPGAs
11:45 AM –
12:15 PM
Ask the experts 1:1 meetings Ask the experts 1:1 meetings Ask the experts 1:1 meetings
12:30 –
1:00 PM
David Ojika (Intel AI Developer Community)
Scientific Deep Learning on Intel® FPGAs
Phillip Schmidt (Intel)
Accelerate Robotics Development with High Precision, Low Power Tracking
Elisabeth Lam (Intel AI Developer Community)
An Animated AI Conference Concierge – Powered by CloudConstable
1:00 –
1:30 PM
Ask the experts 1:1 meetings Ask the experts 1:1 meetings Ask the experts 1:1 meetings
3:30 –
4:00 PM
Jahanzeb Ahmad (Intel): AI+: Combining AI and Other Critical Functions Using Intel FPGAs Yash Akhuri (Intel AI Developer Community)
HadaNets: Flexible Quantization Strategies for Neural Networks
Philip Krejov (Intel)
Deep Learning for VR/AR: Body Tracking with Intel® RealSense™ Technology
4:15 –
4:45 PM
Elisabeth Lam (Intel AI Developer Community)
An Animated AI Conference Concierge – Powered by CloudConstable
David Ojika (Intel AI Developer Community)
Scientific Deep Learning on Intel® FPGA
Phillip Schmidt (Intel)
Accelerate Robotics Development with High Precision, Low Power Tracking

Booth Demonstrations

Model security on FPGA with low latency inference

Add model security to FPGA inference. FPGAs enable developers to create hardware for new or evolving technologies and can process large networks and high-resolution images at low latency.

Accelerate AI segmentation inference in healthcare

Bring quick results to patients by using the Intel® Distribution of OpenVINO™ toolkit to run inference on medical imaging data. With the toolkit you can deploy segmentation AI models on CPU, GPU, VPU, and FPGA or even run heterogeneously for smart device execution.

Efficient and accurate binarized DNNs

Make your deep learning faster with little accuracy loss. Xnor’s efficient binarized deep learning models combined with the Intel® Distribution of OpenVINO™ toolkit can provide accurate real-time person detection at 1000 fps on an Intel® Core™ i5 processor.

Video analytics optimized for Edge AI at Scale

Edge servers can efficiently support deep learning without depending on the cloud. Using the Intel® Distribution of OpenVINO™ toolkit, you can run mixed workloads on this power- and cost-efficient inference platform.

Low power inside-out tracking for robotics, drones and more

With its small form factor and low power consumption, the Intel® RealSense™ Tracking Camera T265 has been designed to give you the tracking performance you want, off-the-shelf. Cross-platform, developer friendly simultaneous localization and mapping for all your robotics, drone and augmented reality rapid prototyping needs.

Applied depth sensing to real world applications

Intel® RealSense™ Stereo depth technology brings 3D to devices and machines that only see 2D today. Diverse capabilities and technologies make Intel® RealSense™ Depth Cameras suitable for a wide range of applications.

Large scale skeletal reconstruction for volumetric rendering

Scaling visual computing across the cloud is challenging and can be expensive. This demo shows a new framework for distributing complex media analytics workloads across many inexpensive cloud compute instances.

Road scene understanding in chaotic driving conditions

Intel’s research collaboration with the International Institute of Information Technology (IIIT), Hyderabad created the India Driving Dataset (IDD), consisting of 50,000 curated images with annotations of 34 object classes, from 182 drive sequences on Indian roads.

Probabilistic computing for human intent prediction

In this demo, we show how a combination of neural networks and approximate Bayesian computation (ABC) can achieve real-time probabilistic inference applied to a human intent prediction scenario. This interactive demo lets users explore and observe the behavior of the predictions.

Compression aware tracking for high density visual cloud analytics

We introduce an ultra-fast tracker, Compression-Aware Tracking (CAT), a method of leveraging the compressed bit-stream’s Motion Vectors (MVs) for faster object-detection. The demo also shows a popular tracking algorithm, Kernelized Correlation Filter (KCF) for comparison.

Intel® AI: In Production

Knowing where to start the development journey is important, because development and deployment of vision applications utilizing deep neural networks can be challenging selecting hardware for inference. Intel’s AI at the Edge interactive kiosk will guide you on how to find information about the Intel® Distribution of OpenVINO™ toolkit, Intel® Neural Compute Stick 2, vision accelerator kits, Intel® Vision Accelerator Design products and Intel’s partner ecosystem for AI at the Edge.

Learn more

Intel® AI Developer Community

Spotlight on community members’ work in Applied Vision Research that utilizes Intel AI Technologies.

  • HadaNets: Flexible quantization strategies for neural networks
  • Conference Connie: The helpful AI assistant
  • Manufacturing failure detection at the edge
  • Automated screening for diabetic retinopathy
  • Heterogeneous computing for scientific problems

Intel Research Datasets from Intel® AI Lab

At Intel, we believe there is a virtuous cycle between datasets, research and compute that’s leading to the tremendous growth we are seeing in AI. At the Intel AI booth at CVPR, we will be demonstrating recent datasets we’ve contributed to the community, jointly with other partners, such as a multi-view satellite imagery dataset (MVOI) and a hierarchical 3D shapes dataset. Learn more.

Intel® Neural Compute Stick 2

Write Once, Deploy Everywhere

It’s now easier to develop applications for heterogeneous execution across the suite of Intel acceleration technologies. Develop once and deploy across Intel CPU, Integrated Graphics, Intel® Neural Compute Stick 2 and Intel® Vision Accelerator Design products. Explore the Intel® Distribution of OpenVINO™ toolkit today.

Learn More

Careers

If you are interested in discovering AI careers that reshape business and society, be sure and stop by our booth and meet our recruiting team or visit the Intel AI careers page where you can explore different roles and join our talent network.

More Ways to Engage

Follow us @IntelAI and @IntelAIResearch for more updates from @CVPR and the Intel AI research team!