Junlin Han

Profile

Greetings, welcome to my website!

I am a researcher at FAIR working on unified pre-training with Dr. Mike Lewis, and a last-year PhD (DPhil) student at the Torr Vision Group, University of Oxford, working with Prof. Philip Torr.
Before that, I worked as a research scientist at Cybever, a 3D AIGC startup for a few months. I obtained my BSc from the Australian National University under the supervision of Dr. Lars Petersson, Prof. Hongdong Li, and Prof. Ian Reid.

News

[03.2026] Releasing a large-scale exploration on unified multimodal pre-training, packed with useful insights into what’s next for pre-training🚀🚀!!
[02.2026] Two papers accepted to CVPR 2026 (Ego-edit and AuralSAM)! Ego-edit introduces all you need (dataset, method, benchmarks) for egocentric editing!
[01.2026] Two papers accepted to ICLR 2026 (Learning to see before seeing with oral presentation and Kaleido)! In LSBS, we show how visual priors emerge from LLM pre-training and demonstrate how we can deliberately leverage them to build more powerful MLLM capabilities.
[11.2025] Joining FAIR to start working with Mike Lewis on unified MLLM pre-training! We pursue the next paradigm of pre-training.
[09.2025] Hallucination at a Glance was accepted to NeurIPS 2025, a new dataset for benchmarking and training MLLMs for fine-grained visual reasoning!
[07.2025] 3D2EP was accepted to SIGGRAPH 2025, a new method for 3D shape decomposition that represents objects as parametric primitives.
[05.2025] Flex3D was accepted to ICML 2025, a new two-stage pipeline for high-quality 3D generation and reconstruction!
[03.2025] One paper on point-cloud segmentation was accepted to CVPR 2025, congrats to Zhaochong!
[12.2024] Our previous work on low-level vision won the best paper award in Remote Sensing!
[12.2024] We will be organizing the 2nd Workshop on Foundation Models in the Wild at ICLR 2025!
[11.2024] Two papers (3D-GPT and DreamBeast) were accepted to 3DV 2025. Congratulations to Chunyi and Runjia! 3D-GPT is pioneering work in large-scale 3D scene generation, and DreamBeast is a fascinating project generating fantastical 3D animals.
[07.2024] VFusion3D and Unicorns were accepted to ECCV 2024! VFusion3D is the first work exploring scalable 3D generative/reconstruction models as a step towards a 3D foundation. Check it out if you are interested in 3D generation.
[04.2024] We are organizing ICML 2024 Foundation Models in the Wild Workshop, submissions on related topics are very welcomed!
[10.2023] Starting my PhD at Meta and Oxford!
[07.2023] Hyperbolic Audio-visual Zero-shot Learning was accepted to ICCV 2023!
[07.2023] Graduated from ANU with First-Class Honours.
[10.2022] Awarded as a top reviewer in NeurIPS 2022!
[07.2022] Blind Image Decomposition (BID) was accepted to ECCV 2022! A new low-level vision task that better adapts to complex real-world scenarios. Check our project page for more.
[05.2022] You Only Cut Once (YOCO) was accepted to ICML 2022! Check here for our work on how to perform data augmentation.
[12.2021] Joined Australian Institute for Machine Learning (AIML), University of Adelaide as a visiting research intern, hosted by Prof. Ian Reid!

Research Interests

My research focuses on multimodal foundation models, including multimodal large language models and generative models, as well as their interactions and integration. You can click below to view some representative papers in each topic.

I approach them from a vision-first perspective, aiming to make vision a central component of general intelligence.

(1) Multimodal foundation models

Selected projects: Visual priors in LLM pre-training, Llama4

(2) Generative (usually 3D/video) models

Selected projects: Scaling-up 3D gen, High-quality 3D asset gen, Scaling up neural rendeing

(3) Integration and interactions

Selected projects: Unfied pre-training, LLM for 3D world modeling

Education

Postgraduate (01.2024 - ongoing)

PhD in Computer Vision, University of Oxford

Prof. Philip Torr

Undergraduate (02.2019 - 07.2023)

B.S. Information Technology (Honours), ANU

Dr. Lars Petersson

Prof. Hongdong Li

Prof. Ian Reid

Research Experience

PhD Student Researcher at Meta (10.2023 – ongoing)

FAIR (12.2025 – ongoing)
Host: Dr. Mike Lewis
Topics: Unfied pre-training
GenAI Llama Vision (08.2024 – 12.2025)
Host: Dr. Filippos Kokkinos
Topics: MLLM post-training and pre-training
GenAI RepliGen (10.2023 – 07.2024)
Host: Dr. Filippos Kokkinos
Topics: 3D Generation, video generation

Lead Research Scientist, Cybever (05.2023 – 09.2023)

Topic: Generative models for 3D world modeling

Visiting Researcher at AIML, University of Adelaide (12.2021 – 05.2023)

Advisor: Prof. Ian Reid
Topics: Data-centric AI, memorability

Research Student at Data61-CSIRO & Australian National University (08.2020 – 05.2023)

Advisors: Dr. Lars Petersson and Prof. Hongdong Li
Topics: Generative models, self-supervised learning

Publications

Recent representative papers are highlighted.

Academic Services

Workshop Organization

Conference review

Journal review

Awards

Meta PhD Scholarship, 2023
for PhD research at University of Oxford
Best paper award, 2022
from Remote Sensing journal
Chancellor's Letter of Commendation, 2022
for best academic performance at Australian National University
Top Reviewer, 2022
NeurIPS (Conference on Neural Information Processing Systems)
Second Best Presentation Award, 2021
AIM (Active Integrated Matter) Conference
Top-up Scholarship, 2021
for research work at Data61-CSIRO
Undergraduate Vacation Scholarship, 2020
for summer research at Data61-CSIRO

Miscs

At the beginning, I was born and raised in Chengdu, which is famous for spice and pandas.
I studied and lived in China, Singapore, Australia, and UK.
Aside from research, I am passionate about traveling (especially nature travel), cooking, gaming, and swimming.