Iro Laina

Iro Laina is a computer vision researcher whose work connects visual perception with language and 3D scene understanding. She coauthored a single-image depth-estimation network in 2016 and an unsupervised image-captioning method in 2019. Her PhD is from TUM, after an MSc in Biomedical Computing there and a Diploma in Electrical and Computer Engineering at NTUA; the ECVA PhD Award recognized her scene-understanding dissertation in 2022. She joined Oxford's Visual Geometry Group in 2020 and became a Departmental Lecturer in Computer Vision in 2023, where she continues work on perception for images, videos and 3D scenes.

Now at: University of Oxford
Works: 63
Citations: 4,723
h-index: 19

Publication numbers from OpenAlex, as of 27 October 2026.

Timeline

2016

Single-image depth prediction

In 2016, she coauthors a fully convolutional residual network to estimate a scene's depth map from one RGB image. The model trains end to end and runs in real time on images or videos.
2019

Unsupervised image captioning

In 2019, she coauthors an image-captioning method that generates scene descriptions without annotated image-caption pairs. It aligns image features with language in a shared latent space and decodes the visual features into captions.
2020

Postdoctoral work at Oxford

She joins Oxford's Visual Geometry Group as a PDRA in 2020.
2022

An ECVA PhD Award

Her PhD (Dr. rer. nat.) is from the Technical University of Munich (TUM), preceded by an MSc in Biomedical Computing there and a Diploma in Electrical and Computer Engineering at the National Technical University of Athens (NTUA). In 2022, the ECVA PhD Award recognizes her dissertation, “Semantics, Language and Geometry: Learning to Understand the Scene,” for its work on scene understanding through geometric, semantic and linguistic components.
Since 2023

Departmental Lecturer at Oxford

She becomes a Departmental Lecturer in Computer Vision at the University of Oxford in 2023. In 2024, she coauthors a NeurIPS paper that trains video segmentation with long-term point trajectories; in 2025 she coauthors “Layered Motion Fusion,” lifting motion segmentation to 3D in egocentric videos, and in 2026 “Mesh4D,” on 4D mesh reconstruction and tracking from monocular video. Her research focuses on perception systems for understanding images, videos and the 3D world through unsupervised and language-supervised learning.