Computer Vision
and Augmented Reality
Conducting Research & Development on various solutions for object recognition and augmented reality areas
Learn more

Our group advances AI technologies for computer vision, deep learning, and 3D navigation. We harnesses it to advance the scientific frontier, as well as build novel industrial and commercial solutions.

Our Research focus areas cover a wide range of topics in Computer Vision, including Augmented Reality, Learning with Limited Labels (Transfer Learning, Few-Shot Learning for detection and recognition, Semi- and Self- supervised learning, Transductive Learning, Cross-domain Few-Shot Learning), multimodal learning, Domain Adaptation, and Efficient AI.

Our past research projects included a touch of medical imaging, retail vision applications, and AI for the environment.

Oded - The manager
Manager Computer Vision and Augmented Reality,
IBM Research - Haifa

Augmented Reality

Our team is developing an augmented reality (AR) enterprise-ready platform in the form of visual operation guidance, aiming to improve the efficiency of technical staff and preserve their knowledge, and by making it reusable over time. To accomplish that, we are enhancing various AR mobile platforms with 3D navigation, object recognition, deep learning as well as home-grown algorithms and tools. Our AR platform can provide easy-to-use, step-by-step procedures that guide technicians of any experience level at the physical site.

Communicating with remote expert using Augmented Reality

IBM Maximo Assist augments asset maintenance with AI by offering our team’s two-ways interactive AR peer guidance assistance solution on a mobile device. The offering has recently been embedded and now being offered as part of Maximo Assist product.

Building Augmented Reality experience

The AR Modeling Studio enables customers to build and maintain their own self-guided AR experiences. With our studio, the traditional tick list is transformed into a live experience, building an AR-enabled experience the technician can follow to perform maintenance, troubleshooting and obtain training.

Searching and downloading Augmented Reality experience

The AR Store is customer’s single go-to-place where they can download and consume their AR experiences. It focuses on bringing them the latest AR experiences for their organization that were published from the AR Modeling Studio, and now made available to be consumed on our IBM Research ARtisan app.

Self-guided troubleshooting with Augmented Reality

IBM Research ARtisan is a mobile app that allows consuming a self-driven, interactive, AR enabled experience. The pre-trained AR model and visual features enabling object recognition, triggering the AR to super impose instructions, links and real time information at specific anchored and tracked spots.

Combining Augmented Reality with IBM Visual Inspection

IBM Maximo Visual Inspection adds intelligent “eyes” to customer’s operations: it enables them to quickly identify defects in production outputs as well as remotely monitor assets for potential disruptions. When brought together with AR, Visual Inspection can be triggered every few steps to ensure things were performed correctly, and thus lead to zero escapes of defects from the manufacturing line.


View by:
Conferences & Journals
Press release

Computer Vision & Deep learning

Automating cafeteria checkout enhanced with food visual recognition

A European company building PoS technology for management of cafeterias and restaurants are leveraging our food recognition algorithms for detecting permanent food and training daily food dishes. Training the daily food dishes is based on our few-shot detection Research work and technologies.

RepMet, CVPR 2019 ->

Learning with Less Labels (LwLL)

The LwLL program makes the process of training ML models more efficient by reducing the amount of labeled data required to build a model by six or more orders of magnitude. It also helps reducing the amount of data required to adapt models to new environments to tens of labeled examples. As part of a competition lead by the DARPA organisation, we are developing Active Learning classification and detection methods for a challenging ensemble of tasks, posed for multiple research and industry teams.

Learn more ->

Analyzing COVID document with Charts Understanding

The document search and question answering in the One Stack Conversion service are extended to include charts data, using our solution for detection and analysis of charts and related objects on document pages, retrieving the underlying tabular data.

Recognition ground cover (HORIZON 2020 EU project)

Geologists can model and predict floods caused by heavy rains to allow local authorities to make better decision on actions leveraging a large-scale ground monitoring solutions co-developed by our team. Our services enables the detection of various types of land coverage and is placed at the heart of a crowd-sourced information gathering system. Scent project.

RepMet, CVPR 2019 ->

Monitoring planogram compliance in Retail

Tracking the misplaced and out of stock items on the supermarket shelves is a tedious task. Our technology helps automate the process of planogram monitoring by detecting and analyzing shelf items among thousands of stock products. The underlying algorithm performs real-time instance recognition using a single training image per object.

CVPR 2017 ->