Sally (Chuhan) Chen 陈楚涵

I am a first year PhD student at CMU Robotics Institute, co-advised by Prof. Matthew O'Toole and Prof. Deva Ramanan. I obtained Masters in Computer Vision at CMU as well.

I did my undergraduate studies in the University of Toronto majoring in Computer Engineering, where I was fortunate to work with Prof. Florian Shkurti on self-driving adversarial attack with differentiable rendering. I also explored research in transit signal priority with Prof. Amer Shalaby and content ranking with ML with Prof. Scott Sanner.

During my undergrad I studied at Kyoto University for one semester supported by JASSO scholarship, during which I worked breifly with Prof. Tatsuya Kawahara on dialogue generation, and also explored many many beautiful architectures and gardens.

Email  /  CV  /  Github

I'm interested in computer graphics, computational photography and vision research with a focus on neural rendering. In terms of application, I'm interested in healthcare, smart cities and creative design tools. Some representative research projects I have been involved in are listed below.

We create differentiable simulators with NeRF where we insert objects in a differentiable way and learn the objects' position and texture (ie. adversarial attacks) that make a self-driving car deviate from its original path.

We model multiple implicit deformation fields with local semantic rig-like control via 3DMM-based parameters and representative facial landmarks, and propose a novel local control loss and attention mask mechanism that renders sharper locally controllable nonlinear deformations than previous implicit monocular approaches, especially mouth interior, asymmetric expressions, and finer skin details.

We proposes a dual-objective two-way Transit Signal Priority algorithm that concurrently optimizes transit delays and reliability (headway adherence) using Deep Reinforcement Learning. Further, the agents are enhanced with a coordination algorithm for an optimized solution balancing opposite directions. The algorithm reacts adaptively to real-time bus performance.

We train multiplane images (MPIs) that estimates depth map of a scene from single-view images with varying focal distances.

Aphasia is a language disorder that causes patients to lose the link between language (what one says) and meaning (what one wants to say). We interviewed speech pathologists for aphasia and created a mobile application that assists aphasia patients with rehabilitation through language exercises that help patients re-establish the connection between words and thooughts.

I'm interested in learning languages & linguistics in general. I speak Mandarin and English fluently, Japanese somewhat fluently (but have forgotten much due to lack of practice) and know some Latin (ahh all the conjugations..). Big thanks to Prof. Atsuko Sakaki - from her class I picked up some Classical Japanese, and now can read some classical text with the help of grammar books. Currently learning Spanish, trying hard to practice tongue-rolling (but do feel occasionally that this might be genetic..) and hopefully can speak it fluently one day! And lastly, I'm from Nanjing, China, so I also speak the Nanjingese dialect, rather sad to see this dialect disappearing and hopefully can contribute to maintaining it.