Shumian Xin

From prototype to production • evaluation • reliability • adoption

I’m a Computer Scientist at Adobe’s Emerging Products (NextCam) team. I build and ship consumer camera and editing experiences end-to-end, spanning both traditional algorithms and generative AI. I’m a core contributor to Project Indigo, an experimental iOS camera app that reached #5 in US Photo & Video and 1M downloads in 4 weeks.

We operate like a startup inside Adobe, and I own features from problem framing and success criteria to hands-on implementation and release quality, then iterate through instrumentation and user feedback. I care about evaluation, reliability, and adoption, and I translate technical detail into clear docs and demos. I’m currently exploring agentic, multimodal UX and how to evaluate it for usefulness, reliability, and trust.

Product training: Stanford Online Product Management Certificate

Previously, I earned my Ph.D. in Robotics from Carnegie Mellon University, advised by Prof. Ioannis Gkioulekas and Prof. Srinivasa Narasimhan. During my Ph.D., I interned on the Google Camera team. My non-line-of-sight imaging work received the CVPR 2019 Best Paper Award.

Email  /  CV  /  Google Scholar  /  LinkedIn

Indigo is an experimental iOS camera app for photography enthusiasts, built to deliver a natural SLR-like look and deeper creative control on mobile.

Highlights include natural HDR/SDR rendering, computational RAW (DNG), pro controls (including computational controls), Lightroom Mobile integration, and a Tech Previews channel (AI Denoise, Remove Reflections).

My role: I own features across capture and editing, translating requirements into design tradeoffs and hands-on algorithm development, integration, tuning, and quality validation. I partner cross-functionally to set the quality, performance, and reliability bar and ship reliably at scale.

Diffusion-based refocus for practical post-capture editing, generating a perceptually consistent focal stack from a single defocused image, with emphasis on controllability, visual consistency, and workflow integration.

Single-shot defocus estimation and deblurring using dual-pixel sensors, jointly estimating a defocus map and reconstructing an all-in-focus image for robust post-capture effects under real-world capture conditions.

A theory of Fermat paths for non-line-of-sight shape reconstruction, enabling high-resolution recovery of occluded objects from indirect light transport measurements, with potential applications in robotics and autonomy, search and rescue, and inspection in obstructed environments.

Template from Jon Barron. Last updated in February 2026.