Menu
Home
GitHub

Zhuoran Zhou

Research Assistant,
University of Washington, IPL Lab
Seattle, WA
E-mail: zhouz47@uw.edu

About me

I received my B.S. degree and M.S. degree from the ECE department of the University of Washington, in 2021 and 2023 respectively.
I'm currently doing research at IPL Lab, UW, advised by Prof.Jenq-Neng Hwang. My main research areas are about Diffusion model and its application in 3D vision. Meanwhile, I'm also focusing on multi-modality feature learning in the relevant areas.
My reseach interests also include generation model, medical image and anything relevant to Computer Vision.
I love solving problems by developing models and algorithms in the area of Computer Vision. Though I was supposed to work as a full-time software engineer in the late 2023, after interning at Doordash and Amazon I decided to pursue a PhD degree instead as I am motivated to explore and work by my own will, as opposed to spending my precious time for the dreams of others.

Research

Current work

  • Huamn Gaussian 3D splatting

  • Unified Motion Generation

Publication

  1. Jiang, Z.*, Zhou, Z.*, Li, L., Chai, W., Yang, C. Y., & Hwang, J. N. (2023). Back to Optimization: Diffusion-based Zero-Shot 3D Human Pose Estimation. arXiv preprint arXiv:2307.03833. (*equal contribution)(WACV 2024)[arxiv] [project page]

  2. Jin, Y., Zhou, Z., Yang, Z., Wang, J., & Hwang. J. N. (2023). Latent Prompting Network for Controllable Radiology Report Generation. (In submission)

  3. Jiang, Z., Chai, W., Li, L., Zhou, Z., & Hwang, J. N.(2023). Towards Unified Human Pose Estimation via Contrastive Learning. (In submission)

  4. Zhou, Z., Jiang, Z., Chai, W., Yang, C., & Hwang. J. N.(2023). Efficient Domain Adaptation via Generative Prior for 3D Infant Pose Estimation. (WACVW 2024)

Education

M.S in Electrical & Computer Engineering,UW, Seattle, WA, 09.2021-03.2023

B.S in Electrical & Computer Engineering,UW, Seattle, WA, 09.2017-06.2021

Research Experience

  1. Research Assistant, Information Processing Lab, UW, Seattle, WA, 11.2022-Present

    • Developed a human pose estimation optimization-based model which integrates a score matching diffusion model for iterative refinement to adjust coarse ray-projected 3D poses in the 2D-3D lifting task. Achieved zeros-shot SOTA performance even compared to deep-learning based models.

    • Transferred the idea of diffusion-refined pose estimation to the infant domain by efficient domain adaptation from the adult priors via a controllable branch and a condition-guided data augmentor.

    • Adopting CLIP’s visual-text understanding to bridge three modalities: point cloud, image and text through the integration of task-oriented prompt tuning in the point cloud downstreams.

  2. Part-Time Microsoft Azure Vision Research Intern, Seattle, WA, 01.2023-05.2023

    • Established a classifier-guided DDPM Diffusion model trained by multiple symptom X-ray images to provide super-class pathology visual prompts to a VQVAE-based latent-prompt medical report generation model fine-tuned from GIT image captioning model and Llama. Disclosed disease-afflicted regions via anomaly map difference between healthy and diseased images generated by the diffusion model.

    • Integrated anomaly map tokens with learnable discrete latent prompts representing sentence-wise GMM pathology clusters to facilitate high-quality medical report generation, resulting in SOTA performance.

  3. Undergrad Research Assistant, Information Processing Lab, UW, Seattle, WA, 08.2020-01.2021

    • Cooperated with NOAA in labeling fish mask data and fine-tuning a Mask-RCNN based model, enhancing its ability of real-time fish species identification and instance segmentation within video footage.

  4. Undergrad Research Assistant, Dr.Gire Lab Lab, UW, Seattle, WA, 06.2019-08.2019

    • Simulated locomotion patterns of rats subjected to varying voltage levels and predicted the positions when rats became occluded using a Kalman Filter based model.

    • Applied the DeepLabCut model to keep track of multiple rats with bounding boxes in real time.

Work Experience

  1. Zongmu - Autonomous Driving Mobile Research Intern, Shanghai, 06.2021-09.2021

    • Tuned a Detectron2 ABCNet model, facilitating recognition of vehicle plates and parking lot numbers with an accuracy of over 85% under severe occlusion and various brightness conditions.

    • Built an MQTT protocol-based network to transmit detection results to a cloud server which may reflect information on the mobile APP in less than 1.5 seconds. Additionally implemented an ONNX versioned model to support local execution on mobile devices.

  2. Telenav - Software Engineer Capstone Intern, Remote, 01.2021-06.2021

    • Co-worked with Telenav Inc. in a group of three to develop a Java library to detect errors in the text queries transcribed from voice, and re-rank the queries by evaluating TF-IDF and n-gram correctness.

    • Designed and developed an Android App to deploy the testing library so that the new voice recognition system could tolerate accents and noisy environments.