Opening and Welcome

Keynote: How Far Are We From the Restore Any lmage Model (RAIM)? [online]

• [Abstract]
• [Slides]
• [Zoom] (pwd: 54892391)

Image restoration and enhancement is a widely studied problem in low-level vision. Researchers have proposed a variety of theories and algorithms in order to reconstruct the desired images stably under different imaging conditions. However, we still encounter many challenging issues in practical photography scenarios. The rapid development of deep learning techniques has significantly improved the performance of computational photography; nonetheless, most of the trained deep image restoration models still lack enough generalization capability. Large AI models, especially the pre-trained generative diffusion models, shed light on our dream to develop Restore Any Image Models (RIAM). How far are we from RAIM? Can generative models address the challenges of real-world photography in complex scenarios? In this talk, we will introduce some recent advances along this line with some preliminary discussions.

Lei Zhang

The Hong Kong Polytechnic University & OPPO Research Institute

Keynote: Revisiting Image White Balancing [online]

• [Abstract]
• [Slides]
• [Zoom] (pwd: 54892391)

White balance correction is an essential process for accurate color reproduction in digital imaging. Most modern cameras perform white-balance correction early in the camera ISP, where global white-balance gains are estimated from the captured scene and then applied to the raw image. In this talk, I will discuss the challenges of using learning-based methods for in-camera white balance correction and our work to mitigate these challenges. Additionally, I will address the complexities of correcting white balance in the post-capture stage, where the camera ISP has already applied photofinishing steps to produce the final image. I will present our work on post-capture white-balance correction and editing, and then cover our prior research in spatial white-balance correction, which goes beyond the traditional global white balance correction applied onboard cameras. Lastly, I will discuss the impact of white-balance correction on other downstream computer vision tasks.

Mahmoud Afifi

Camera Software Engineer, Google

Keynote: Towards A Better Camera in the App and AR Glass

• [Abstract]
• [Slides]

With 422 million daily active users (DAUs, Q1 2024), Snapchat's camera stands as one of the most widely utilized cameras today. Snapchat users rely on this camera to capture moments, express themselves, communicate visually, and enjoy shared experiences. We empower the camera to cater to these diverse needs through our cutting-edge camera technology. In this presentation, I will unveil a series of our endeavors aimed at enhancing the camera's capabilities. Firstly, I will discuss our efforts to improve the in-app camera's image quality. Secondly, I will introduce how we make a fun camera through the special filters. Thirdly, I will explore how we have integrated additional hardware to create superior cameras for the next-generation AR glass. Finally, I will touch upon our innovations in transparent displays and transparent cameras.

Jian Wang

Staff Research Scientist, Snap Research

Keynote: Passive Ultra-Wideband Single-Photon Imaging

• [Abstract]

High-speed light sources, fast cameras, and depth sensors have made it possible to image dynamic phenomena occurring in ever smaller time intervals with the help of actively-controlled light sources and synchronization. Unfortunately, while these techniques do capture ultrafast events, they cannot simultaneously capture slower ones too.

I will discuss our recent work on passive ultra-wideband single-photon imaging, where we consider the problem of imaging a dynamic scene over an extreme range of timescales simultaneously from seconds to picoseconds completely passively, without any light sources under our control, no synchronization, and not much light. We call this new imaging regime passive ultra-wideband imaging and show that this regime enables post capture refocusing in time—from transient to everyday timescales and can be captured using single-photon cameras. We show the potential of passive ultra-wideband imaging by experimentally demonstrating several never-seen-before abilities, including recording ultra-wideband video, which can be played back later at 30 Hz to show everyday motions—but can also be played a billion times slower to show the propagation of light itself.

Mian Wei

PhD Student, University of Toronto

Closing