IQI Lab
Intelligent Quantum Imaging
ShanghaiTech University · Shanghai
We co-design the optics, the illumination, and the algorithm as one instrument: pulsed LiDAR, indirect time-of-flight, and FMCW radar shrunk to the volume and power budget an agent can carry — and simulated end to end before anything is machined.
Active sensing is a single design problem stretched across a lot of disciplines. We work on all of it: the emitter and detector that produce the measurement, the simulation that predicts it, the reconstruction that inverts it, and the agent that decides where to point next.
Hardware
SPAD arrays, VCSEL illumination, and mm-wave front ends packed into a volume an agent can actually carry. The constraint that matters is not resolution on a bench — it is millimetre depth inside a few cubic centimetres and a few hundred milliwatts.
Closed loop
An agent that moves can choose where to point, how long to integrate, and how to spend a fixed photon budget. We treat sensing as an action: the pointing policy and the reconstruction are learned against each other rather than bolted together afterwards.
Differentiable
One differentiable chain from source to readout: emitter waveform, lens prescription, scene light transport, detector jitter, and the reconstruction network. Gradients reach the glass, so the lens and the algorithm stop being separate deliverables.
Inverse problem
Transient transformers, neural transient fields, and masked pretraining that recover geometry from measurements too sparse or too noisy for classical inversion — including light that never took a direct path to the sensor.
2026
FermatFormer is accepted to IEEE TPAMI.
ICCP 2026 Best Paper Honorable Mention
2026
GenPIE is accepted to SIGGRAPH 2026 and appears in ACM Transactions on Graphics.
Lab members are set in bold. Everything below is peer-reviewed or on arXiv; follow a title for the paper, code, or project page.
2026
FermatFormer: A Fermat Optics Based Neural Architecture for Non-line-of-sight ImagingICCP 2026 Best Paper Honorable Mention
IEEE TPAMI
2026
GenPIE: A Time-Resolved Plenoptic ImagerSIGGRAPH · ACM TOG
2025
TransiT: Transient Transformer for Non-line-of-sight VideographyICCV
2025
MARMOT: Masked Autoencoder for Modeling Transient ImagingarXiv
2025
HOLI-1-to-3: Transient-Enhanced Holistic Image-to-3D GenerationIEEE TPAMI
2023
Enhancing Non-line-of-sight Imaging via Learnable Inverse Kernel and Attention MechanismsICCV
2023
Neural Reconstruction through Scattering Media with Forward and Backward LossesICCP
2023
High-resolution Tomographic Reconstruction of Optical Absorbance through Scattering Media Using Neural FieldsarXiv
2022
Onsite Non-line-of-sight Imaging via Online CalibrationIEEE Photonics Journal
2021
Non-line-of-sight Imaging via Neural Transient FieldsIEEE TPAMI
Principal investigator
ShanghaiTech University
Computational photography, light transport, and 3D vision. Sets the lab's direction across optics, sensors, and learning.
Collaborator
KAUST
Computational imaging and displays, and end-to-end optical design. Joint work on differentiable simulation of the full capture chain.
Faculty
ShanghaiTech University
Transient and non-line-of-sight imaging systems. Leads the lab's instrumentation and time-resolved capture work.
School of Information Science and Technology, ShanghaiTech University, Pudong, Shanghai.
The lab runs its own optical bench: pulsed lasers, SPAD arrays, gated sensors, and mm-wave front ends.
PhD, master's, and undergraduate research positions open year-round. Optics, RF, embedded systems, differentiable rendering, and machine learning all have a seat at this bench — tell us which part of the chain you want to own, and what you have already built.