Spatio-temporal action detection
Localizing who does what, and when, efficiently across long videos.
NewOral at ECCVW 2026
PhD candidate · University of Victoria
Ali Soltaninezhad
Computer vision that works in the real world.
Everything a model sees, and the small part worth a person’s attention.
Research
Perception that holds up outside the lab.
Five threads, one goal: video models that are fast enough, light enough, and robust enough to leave the cluster.
Video understanding
Motion-aware models that reason about events, not just frames.
Privacy-preserving perception
Recognizing actions while protecting the identity of the people in frame.
Environmental monitoring AI
Deployable perception for maritime and ecological monitoring.
Medical image processing
Robust analysis pipelines for clinical and biomedical imaging.
About
Machine perception that survives the messy parts.
I'm a PhD candidate in Electrical and Computer Engineering at the University of Victoria, in Dr. Alexandra Branzan Albu's Computer Vision Lab. I teach machines to read what people are doing in video, fast enough and light enough to run outside a research cluster. In practice that means spatio-temporal action detection, motion-aware video understanding, privacy-preserving perception, and models built to hold up in the wild.
The work has left the lab. Three first-author papers, all of them oral presentations: ECCVW 2026 (the 2nd Marine Vision Workshop at ECCV), ICPR 2026 and CRV 2025. It also powers Archipelago Marine Research's FishVue platform, which took the 2026 VIATEC Innovative Excellence Award. I teach, too: I'm the instructor for UVic's Medical Image Processing course (ECE 435 / BME 403) this winter, on a $10,000 President's Fellowship in Research-Enriched Teaching.
What I care about is computer vision that survives the messy parts: the open sea, bad light, occlusion, motion blur. Not just a clean benchmark. If that sounds like your problem too, let's talk.
Selected projects
Built to run where it's hard.
From reviewer guidance to lightweight action detection and privacy on working vessels.
Loading projects…
Publications
First-author at ECCVW, ICPR and CRV.
Peer-reviewed work on action detection, at-sea monitoring, and video understanding. The three first-author papers were all oral presentations.
- Loading publications…
Teaching
Instructor for Medical Image Processing at UVic.
Jan – Apr 2026
Course instructor
ECE 435 / BME 403, Medical Image Processing. University of Victoria.
2024 – 2026
Super teaching assistant
ECE 471, Computer Vision (2024) · ECE 527, Applied Data Analysis (2025, 2026).
2023 – 2026
Lab instructor and teaching assistant
Control Systems, Digital Design, VLSI Systems, Electronic Circuits, Signals & Systems. University of Victoria.
Honors & awards
Fellowship and national rankings.
$10K
President's Fellowship
Research-Enriched Teaching, 2025.
Top 1%
National BSc entrance exam
Ranked in the top 1% nationally.
Top 3%
National MSc entrance exam
Ranked in the top 3% nationally.
Industry impact
Running at sea. Not on a benchmark.
My research feeds Archipelago Marine Research's FishVue, the platform that automates catch and compliance monitoring on commercial fishing vessels. It helped FishVue win the 2026 VIATEC Innovative Excellence Award.
Read about the award
TrapCounter output on deck cameras: detection, tracking and optical flow counting active traps.
Contact
Let's build something that works.
Working on video understanding, deployable computer vision, or perception in messy real-world conditions? Email is the fastest way to reach me. I read every note.
alisoltaninezhad@uvic.caVictoria, BC, Canada