Motion, Language, and Multimodal Integration for Video Understanding

Mona Ahmadian
PhD in Video Understanding, University of Surrey, UK
January 2026
Thesis PDF LinkedIn
Tweet Share

TL;DR: This PhD in Video Understanding develops a unified framework for understanding complex actions and events in untrimmed, real-world videos by:

  • MOFO: motion-focused self-supervision that explicitly models motion dynamics and motion-sensitive regions in videos.
  • FILS: language-grounded self-supervision that predicts video features in a semantic language space.
  • DEL: supervised multimodal dense event localisation for overlapping and asynchronous audio–visual events.
  • Together, these contributions advance interpretable, motion-aware, and semantically grounded video understanding in the context of digital media.

Downloads

Thesis
PhD Thesis PDF
Associated Papers
MOFO Paper (NeurIPS SSL Workshop 2023)
FILS Paper (BMVC 2024)
DEL Paper (ICCV MMFM4 Workshop / under review at Multimedia 2026)

For project pages with additional figures, videos, and implementation details, see: MOFO, FILS, DEL.

Publications Linked to This Thesis

The core chapters of this thesis correspond to the following peer-reviewed publications:

  1. Chapter 3 – MOFO
    M. Ahmadian, F. Guerin, and A. Gilbert.
    “MOFO: MOtion FOcused Self-Supervision for Video Understanding,”
    In Proceedings of the Self-Supervised Learning Workshop – Theory and Practice (NeurIPS), 2023.
    Project page | Paper (PDF)
  2. Chapter 4 – FILS
    M. Ahmadian, F. Guerin, and A. Gilbert.
    “FILS: Self-Supervised Video Feature Prediction in Semantic Language Space,”
    In Proceedings of the 35th British Machine Vision Conference (BMVC), 2024.
    Project page | Paper (PDF)
  3. Chapter 5 – DEL
    M. Ahmadian, A. Shirian, F. Guerin, and A. Gilbert.
    “DEL: Dense Event Localisation for Multi-modal Audio-Visual Understanding,”
    In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshop on What is Next in Multimodal Foundation Models (MMFM4), and under review at Multimedia, 2026.
    Project page | Paper (PDF)

These works are integrated and extended within the thesis, which situates them in a broader framework for motion, language, and multimodal integration in video understanding.

BibTeX

@phdthesis{Ahmadian2026Thesis,
  author    = {Mona Ahmadian},
  title     = {Motion, Language, and Multimodal Integration for Video Understanding},
  school    = {University of Surrey},
  year      = {2026},
  month     = {January},
  address   = {Guildford, United Kingdom},
  note      = {PhD in Video Understanding. Chapters based on MOFO (NeurIPS SSL Workshop 2023), FILS (BMVC 2024), and DEL (ICCV MMFM4 Workshop / under review at Multimedia 2026)}
}