Computer Engineering Student & Undergraduate Researcher
My research focuses on computer vision and multimodal learning, with a particular interest in recognizing individual objects and estimating their physical properties from images through structured, chain-of-thought reasoning. My current work applies these ideas to dietary assessment: identifying the foods in a meal, estimating their portions, and connecting them to nutritional reference data. I develop methods that verify food identities and image regions, recover overlooked items, and address errors that propagate from recognition to nutrition estimation. I am also exploring learning-based approaches that use food appearance, shape, and spatial relationships to improve portion and nutrient estimation across diverse cuisines from the United States, Europe, Oceania, and Asia.
I am Jingbo (Jackie) Yue, a Computer Engineering student at Purdue University. I conduct undergraduate research in the Video and Image Processing Laboratory (VIPER), advised by Prof. Fengqing Zhu.
My recent research examines the reliability of image-based dietary assessment: how food regions can be verified using semantic and geometric evidence, how missed items can be recovered through targeted segmentation review, and how identified foods can be grounded to nutritional reference data.
I previously conducted computer vision research on image-based plant phenotyping and herbicide-injury assessment at Purdue's ABE Plant Sensor Lab (LeafSpec), led by Prof. Jian Jin. Beyond these applications, I enjoy connecting algorithmic image analysis with practical systems engineering.
How can segmentation and spatial consistency help distinguish visible food items from false detections and recover missed items? This direction studies region-level evidence and object-centric recognition.
How can multimodal models check each proposed food item against the image? Our food-item verification work uses region checks and a targeted review of unresolved visual evidence, followed by independent re-verification of recovered items.
How do food identification, portion estimates, and reference-data selection affect the final nutrition estimate? Our Open-KNEAD work studies item-level retrieval from FNDDS and auditable portion estimation.
Jingbo Yue, Bruce Coburn, Jinge Ma, Jui-Feng Chi, Fengqing Zhu
arXiv:2610.11144 · Submitted to IEEE ICASSP 2027
In brief: Verifies proposed food identities and image regions, then uses one targeted recovery pass for overlooked items before nutrition estimation.
Bruce Coburn, Jingbo Yue, Jinge Ma, Siddeshwar Raghavan, Gautham Vinod, Fengqing Zhu
Accepted to IEEE/CVF Winter Conference on Applications of Computer Vision (WACV 2027) · arXiv:2607.12911
In brief: Uses item-wise decomposition and selective FNDDS retrieval to produce an auditable nutrition estimate from a single meal image.
Team research project behind the food-item verification paper: checks item identity and whether image regions support portion estimation, then reviews unresolved areas for missed foods.
Team research project behind Open-KNEAD: decomposes meals into food items and uses selective FNDDS retrieval to support traceable portion and nutrition estimates.
Senior design prototype linking ESP32 controllers, an interactive touchscreen, NFC interactions, and UART communication, with a classical image-processing approach for board-state recognition.
Course project exploring image-based plant phenotyping and herbicide-injury assessment. The public repository contains Jupyter notebook experiments and final presentation materials.
Morphology · Edge detection · HSV color space
HOG descriptors · ROI extraction · Image features
Support Vector Machines · Classification
Homography · Perspective transformation · Mapping
Python · PyTorch · OpenCV · Jupyter
Linux · Git · CUDA workflows
C / C++ · ESP32 · UART · Microcontrollers
Dataset curation · Error analysis · Validation
Data structures · Digital signal processing
Research collaborations, academic inquiries, and professional opportunities.