Niki Amini-Naieni
I am a computer vision and AI researcher, Supernumerary Fellow, and member of the faculty of Computer Science at New College, Oxford. There, I am working on developing a multi-task model for visual understanding of the open world, and bringing this AI system to real-world impact across a diverse range of scientific and industrial applications. My research involves developing and leveraging foundation models, multi-modal vision-language models, and other deep learning models. I am best known for my work on open-world object counting in images and videos. I am currently interning at Google, DeepMind.
Google Scholar · GitHub · LinkedIn · CV
Research
Research themes
- Open-world object counting in images and videos
- Foundation models
- Multi-modal vision-language models
- AI for science and real-world applications
Current focus
Developing a deep learning model that automatically counts, segments, and identifies any type of object in images and videos. Given an image or video and a flexible multi-modal 'prompt' specifying a type of object, the AI system will output segmentation masks, object IDs, frame-level counts, and a global count indicating the number of unique objects that match the prompt. This system should efficiently scale to hundreds and thousands of objects and significanthly accelerate the development of innovations and discoveries across many diverse fields.
Selected Publications
Teaching
I teach computer science tutorials to undergraduates at New College (University of Oxford), lecture in computer vision and deep learning for doctoral students, and have held research-focused tutorials for postdoctoral fellows interested in applying AI to their domains.
- Undergraduate tutorials in Computer Science, New College, Oxford - Supernumerary Fellow (incoming)
- Advanced Deep Learning Course, Intelligent Earth Centre for Doctoral Training, University of Oxford — Co-Lead (2024 - present)
- Computer Vision Course, Autonomous Intelligent Machines & Systems Centre for Doctoral Training, University of Oxford — Teaching Assistant (2025 - present)
- Deep Learning Course, Autonomous Intelligent Machines & Systems Centre for Doctoral Training, University of Oxford — Teaching Assistant (2025 - present)
- Research tutorials on multi-modal foundation models, Schmidt AI in Science Postdoctoral Fellowship, University of Oxford - Tutor (2024 - 2025)
- Example practical I developed with Andrew Zisserman and Iro Laina
Software & Datasets
- CountGD++ — model and code for paper "CountGD++: Generalized Prompting for Open-World Counting."
- CountVid, VideoCount — model, code, and dataset for paper "Open-World Object Counting in Videos."
- CountGD — model and code for paper "CountGD: Multi-Modal Open-World Counting."
- CountGD HF App — Hugging Face App I developed for CountGD from the paper "CountGD: Multi-Modal Open-World Counting."
- ZapCount App — use cases for and app built on top of CountGD from the paper "CountGD: Multi-Modal Open-World Counting."
- Annolid — data annotation tool that uses CountGD from the paper "CountGD: Multi-Modal Open-World Counting."
- InstantCalibration — code for paper "Instant Uncertainty Calibration of NeRFs Using a Meta-Calibrator."
Talks
Recorded talks (see CV for list of both recorded and non-recorded talks)
Prospective Supervision & Collaborations
I am happy to explore potential collaborations and to co-supervise PhD, masters, or undergraduate students in topics relevant to my research agenda. I am also always interested in learning about applicatios of (or potential applications of) my work. Please reach out if you think there is a promising connection.
CV
Contact
niki.amini-naieni@eng.ox.ac.uk
New College, University of Oxford