Girmaw Abebe Tadesse

Girmaw Abebe Tadesse

Kenya
8K followers 500+ connections

About

A Principal Research Scientist and Manager at Microsoft AI for Good Research Lab…

Activity

Join now to see all activity

Experience

  • Microsoft Graphic

    Microsoft

    AI for Good Research Lab

  • -

  • -

  • -

    Oxford, United Kingdom

  • -

    Oxford, United Kingdom

  • -

    London, United Kingdom, Barcelona, Spain

  • -

    London, United Kingdom

  • -

    Barcelona Area, Spain

  • -

    Ethiopia

  • -

    Belgium

  • -

    Lisbon Area, Portugal

  • -

    Ethiopia

  • -

    Ethiopia

Education

Licenses & Certifications

Volunteer Experience

  • STEMNET Graphic

    Ambassador

    STEMNET

    - Present 9 years 6 months

    Science and Technology

  • Udacity Graphic

    Assistant for Machine Learning Workshops, Udactiy London

    Udacity

    - Present 9 years 2 months

    Education

    Assisting attendees of Machine Learning workshops organized by Udacity London team to help its students of the Nano-degree programme in Machine Learning.

Publications

  • A first-person vision dataset of office activities

    In MPRSS'18: 24th International Conference on Pattern Recognition, Beijing, China

    We present a multi-subject rst-person vision dataset of offi ce activities. The dataset contains the highest number of subjects and activities compared to existing office activity datasets. Once activities include person-to-person interactions, such as chatting and handshaking, person-to-object interactions, such as using a computer or a whiteboard, as well as generic activities such as walking. The videos in the dataset present a number of challenges that, in addition to intra-class…

    We present a multi-subject rst-person vision dataset of offi ce activities. The dataset contains the highest number of subjects and activities compared to existing office activity datasets. Once activities include person-to-person interactions, such as chatting and handshaking, person-to-object interactions, such as using a computer or a whiteboard, as well as generic activities such as walking. The videos in the dataset present a number of challenges that, in addition to intra-class differences and inter-class similarities, include frames with illumination changes, motion blur, and lack of texture. Moreover, we present and discuss state-of-the-art features extracted from the dataset and baseline activity recognition results with a number of existing methods. The dataset is provided along with its annotation and the extracted features.

    Other authors
    • Andreu Catala
    • Andrea Cavallaro
  • Visual features for ego-centric activity recognition: A survey

    In WearSys’18: 4th ACM Workshop on wearable systems and applications , Munich, Germany

    Wearable cameras, which are becoming common mobile sensing platforms to capture the environment surrounding a person, can also be used to infer activities of the wearer. In this paper we critically discuss features for ego-centric activity recognition using videos. These features can be learned from data or designed to effectively encode motion magnitude, direction and other dynamics. Features can be derived from optical flow, from the displacement of key-points or the intensity centroid. We…

    Wearable cameras, which are becoming common mobile sensing platforms to capture the environment surrounding a person, can also be used to infer activities of the wearer. In this paper we critically discuss features for ego-centric activity recognition using videos. These features can be learned from data or designed to effectively encode motion magnitude, direction and other dynamics. Features can be derived from optical flow, from the displacement of key-points or the intensity centroid. We also discuss how features are effectively filtered and fused for specific tasks. Features presented in this paper can also be applied to other wearable systems that use accelerometer and gyroscope data.

  • Hierarchical modeling for first-person vision activity recognition

    Neurocomputing

    We propose a multi-layer framework to recognize ego-centric activities from a wearable camera. We model the activities of interest as hierarchy based on low-level feature groups. These feature groups encode motion magnitude, direction and variation of intra-frame appearance descriptors. Then we exploit the temporal relationships among activities to extract a high-level feature that accumulates and weights past information. Finally, we define a confidence score to temporally smooth the…

    We propose a multi-layer framework to recognize ego-centric activities from a wearable camera. We model the activities of interest as hierarchy based on low-level feature groups. These feature groups encode motion magnitude, direction and variation of intra-frame appearance descriptors. Then we exploit the temporal relationships among activities to extract a high-level feature that accumulates and weights past information. Finally, we define a confidence score to temporally smooth the classification decision. The results across multiple public datasets show that the proposed framework outperforms state-of-the-art approaches, e.g. with at least 8% improvement in precision and recall on a 15-hour public dataset with six locomotive activities.

    See publication
  • A long short-term memory convolutional neural network for first-person vision activity recognition

    The IEEE International Conference on Computer Vision (ICCV)

    Temporal information is the main source of discriminating characteristics for the recognition of proprioceptive activities in first-person vision. We propose a novel motion representation that uses a multi-channel stacked spectrograms in order to learn high-level global motion dynamics using convolutional neural network (CNN) with 2D convolutions. The spectrograms are generated from mean grid-optical flow vectors and the displacement vectors of the intensity centroid in a video sample, a window…

    Temporal information is the main source of discriminating characteristics for the recognition of proprioceptive activities in first-person vision. We propose a novel motion representation that uses a multi-channel stacked spectrograms in order to learn high-level global motion dynamics using convolutional neural network (CNN) with 2D convolutions. The spectrograms are generated from mean grid-optical flow vectors and the displacement vectors of the intensity centroid in a video sample, a window of frames that contains the minimum temporal information for classification. We employ a long short-term memory (LSTM) network to encode the temporal dependency among consecutive samples recursively. Experimental results show that the proposed approach outperforms state-of-the-art methods in the largest first-person vision datasets.

    See publication
  • Inertial-Vision: cross-domain knowledge transfer for wearable sensors

    The IEEE International Conference on Computer Vision (ICCV)

    First-person proprioceptive activity recognition infers the activities of a subject from egocentric data. Inertial measurement units (IMU) and wearable cameras are common sensors to collect egocentric data. IMU-based approaches often employ a cascade of hand-crafted features from triaxial motion representation. First-person vision (FPV) approaches generally employ global motion features. Vision-based approaches offer transfer learning capability from pre-trained image models, whereas IMU…

    First-person proprioceptive activity recognition infers the activities of a subject from egocentric data. Inertial measurement units (IMU) and wearable cameras are common sensors to collect egocentric data. IMU-based approaches often employ a cascade of hand-crafted features from triaxial motion representation. First-person vision (FPV) approaches generally employ global motion features. Vision-based approaches offer transfer learning capability from pre-trained image models, whereas IMU provides simplified motion representation. Hence, we propose the transfer of the merits between the inertial and visual approaches for effective recognition of human activities as existing deep frameworks for inertial data are often built from scratch with limited training data. Particularly, we propose sparsity weighted combination of information from different motion streams of IMU and/or FPV. We validate the proposed framework on multiple visual and inertial datasets.

    See publication
  • Robust multi-dimensional motion features for first-person vision activity recognition

    Computer Vision and Image Understanding (CVIU)

    We propose robust multi-dimensional motion features for human activity recognition from first-person videos. The proposed features encode information about motion magnitude, direction and variation, and combine them with virtual inertial data generated from the video itself. The use of grid flow representation, per-frame normalization and temporal feature accumulation enhances the robustness of our new representation. Results on multiple datasets demonstrate that the proposed feature…

    We propose robust multi-dimensional motion features for human activity recognition from first-person videos. The proposed features encode information about motion magnitude, direction and variation, and combine them with virtual inertial data generated from the video itself. The use of grid flow representation, per-frame normalization and temporal feature accumulation enhances the robustness of our new representation. Results on multiple datasets demonstrate that the proposed feature representation outperforms existing motion features, and importantly it does so independently of the classifier. Moreover, the proposed multi-dimensional motion features are general enough to make them suitable for vision tasks beyond those related to wearable cameras.

    Other authors
    • Andrea Cavallaro
    • Xavier Parra
    See publication

Honors & Awards

  • Erasmus Mundus Double Doctorate Fellowship - Interactive and Cognitive Environments

    -

Languages

  • English

    Full professional proficiency

  • Spanish (A2)

    -

  • Italian (A1)

    -

  • Amharic

    Native or bilingual proficiency

More activity by Girmaw Abebe

View Girmaw Abebe’s full profile

  • See who you know in common
  • Get introduced
  • Contact Girmaw Abebe directly
Join to view full profile

Other similar profiles

Explore collaborative articles

We’re unlocking community knowledge in a new way. Experts add insights directly into each article, started with the help of AI.

Explore More

Add new skills with these courses