Qiuyang

On-device saliency and smart cropping for iOS images in Swift.

• Applied Apple’s iOS Edge Saliency Model within the Vision framework, backed by the Neural Engine, to generate bounding boxes for salient regions and smart-crop images in Swift.
• Assessed model performance on more than 1.5k images across diverse datasets, including selfies, food, and landscapes.
• Built a visual prototype in Xcode showing the real-time effect of smart cropping, with side-by-side comparison against the baseline model.
• Compared the iOS Edge Saliency Model, the baseline model, and Amazon’s in-house saliency model on latency and IoU, with suggestions for future improvements.
• Shipped the beta build to TestFlight so the feature stayed available in later builds.

• Swift
• iOS Vision
• Neural Engine
• Xcode
• TestFlight

Human body segmentation in video for 3D body reconstruction. Advised by Prof. Henry Fuchs, with Qian Zhang.

• Trained a ResNet body-segmentation model on human body images as a prior for a 3D body reconstruction project.
• Used temporal information through a propagation module and optical flow, improving video segmentation accuracy and overall IoU.

• ResNet
• Optical flow
• Video segmentation

Java Spring Boot APIs and test suites.

• Integrated 20+ APIs with Java Spring Boot, MyBatis, MySQL, and Postman.
• Wrote test suites with Mockito, reaching over 80% coverage across features.

• Java
• Spring Boot
• MyBatis
• MySQL
• Mockito