Munawar Hayat

Timeline

1. 2009–2015

   From engineering to a UWA PhD

   He earns a BSc in engineering at the National University of Sciences & Technology in 2009 and a master's in space science at Luleå Tekniska Universitet in 2011. He completes a PhD in computer science at the University of Western Australia in 2015.

2. 2015

   The Robert Street Prize

   His University of Western Australia thesis, Learning Techniques for Multi-Modal Facial Analysis, wins the Robert Street Prize in 2015. UWA awards the prize for the most outstanding doctoral thesis each year.

3. 2020–2021

   Image restoration with CycleISP and MIRNet

   In 2020, he coauthors CycleISP, which models camera imaging pipelines to generate synthetic training data for image denoising, at CVPR and MIRNet, which keeps high-resolution representations while drawing on low-resolution context, at ECCV. Multi-Stage Progressive Image Restoration follows at CVPR 2021.

4. 2021

   Transformers in Vision survey

   He coauthors Transformers in Vision: A Survey with Salman Khan, Muzammal Naseer, Syed Waqas Zamir, Fahad Shahbaz Khan and Mubarak Shah, posted in January 2021. The survey covers self-attention, large-scale pre-training and applications from recognition to 3D analysis, and ACM Computing Surveys accepts it in December 2021.

5. 2021–2023

   ARC DECRA fellow at Monash

   At Monash University in Melbourne, he is a Senior Lecturer in the Faculty of IT and holds an ARC DECRA Fellowship for 2021–2023, worth $425,613.

6. 2022

   Restormer at CVPR 2022

   He coauthors Restormer with Syed Waqas Zamir, Aditya Arora, Salman Khan, Fahad Shahbaz Khan and Ming-Hsuan Yang, accepted at CVPR 2022. The transformer captures long-range pixel interactions in high-resolution images and reaches state-of-the-art results on deraining, motion and defocus deblurring, and denoising.

7. 2025

   Multimodal papers at NeurIPS 2025

   At NeurIPS 2025, he and his Qualcomm AI Research team present papers on object hallucination in vision-language models and on composed retrieval with combined text and image queries. A third, MultiHuman-Testbench, is a benchmark of 1,800 samples with 5,550 unique face images for generating images of multiple humans.

8. As of October 2026

   Multimodal AI team at Qualcomm

   He is a Principal AI Researcher at Qualcomm, where he leads the Multimodal AI team at Qualcomm AI Research. His work focuses on efficient multimodal generative AI models and perception systems for devices including mobile, AR/VR, IoT and autonomous vehicles.