Rongxiang Zhang

Main Achievements: Implemented multi-modal LLM capabilities in TikTok ads, enabling game name recognition, ad title evaluation, and content consumption prediction. Led Capcut series image quality enhancements, generating >100M RMB commercial revenue. Developed SD-based super-resolution/color algorithms surpassing Magnific.ai benchmarks. Team & Project Management: Led cross-business image quality support for Capcut, Toutiao, Douyin, TikTok across shooting/editing/marketing/live streaming scenarios. Drove AIGC+image quality R&D roadmap, maintained tech leadership, and enhanced team influence. Spearheaded VLM applications in image-text understanding, driving tech iteration and business metrics growth. Technical R&D: Ad AI System: Built GPT-4O/Gimni2.5pro Teacher-Student framework with AutoPrompt optimization. Achieved 15%+ recall gain in game name prediction and 79% accuracy in ad title scoring. Implemented multi-modal CTR prediction for micro-headline CTR optimization. Content Moderation: Created hybrid frame extraction (scene changes/clustering/entropy) achieving ~99% ANSA recall. Deployed CLIP+TSN+MLP model for low-quality video detection. AIGC Super-Res: Engineered SD V1.5/XL-based pipeline with controlnet/diffBIR/StabelSR/MGLDVSR/BasicVSR++ architectures. Achieved superior ID consistency, detail clarity, and temporal stability. "One-Click Movie": Delivered portrait segmentation (DepthAnythingV2), depth prediction, bokeh (BokehMe), and tone optimization. Drove 4% paid UV penetration increase and ~20K/month GMV growth. AI Imaging Pipeline: Developed super-resolution, old photo restoration, document enhancement, auto-color calibration, and intelligent HD algorithms. Deployed aesthetic/facial scoring systems and multiple successful APP products.

Main Achievements: Led industry-first motion capture development for vivo X50~X90 series (v1.0-5.0), earning 2020 Innovation Award Pioneered multi-exposure fusion for noise reduction/deblurring/wide dynamic range imaging Delivered industry's first Zeiss-certified handheld AllInFocus depth extension technology Team & Project Management: Directed technical roadmap for X50~X90 motion capture systems Coordinated cross-location R&D teams to accelerate project delivery Established partnerships with component/algorithm vendors (EVS, 3rd parties) for advanced tech integration Built CNN/GAN training/inference framework improving R&D efficiency Technical R&D: Motion 3A Control: Implemented intensity measurement via Motion Metering, 17KPs skeletal tracking, and PWCNet optical flow. Enabled precise AE/AF control for moving subjects. Local BLC/AWB reduced color bias in ultra-high ISO dark scenes. Multi-frame Fusion: Developed FBA/RawEnhance 1.0/2.0 algorithms in YUV/Raw domains. Combined exposure bracketing, DCG sensor dual-stream, and noise modeling. Achieved 30%+ SNR improvement via Quad/Binning Bayer fusion. Motion Deblurring: Created motion direction/intensity equalization tech. Deployed MS-HINET architecture for 360° directional deblurring (0-32px). Solved exposure-speed mismatch issues. AllInFocus: Automated depth scanning with MF-UNet fusion. Produced motion-robust full-depth synthesis. Enabled single-frame optical defocus correction via MS-HINET. Training Framework: Built IQT-DL architecture with sample cleaning/augmentation/balancing. Introduced GTEnhance and motion degradation pipeline. Accelerated model development by 40%.

Face Dectection and Alignment Face Makeup / Scene Recognition Main Achievements: Built the R&D team in Hangzhou from scratch, smoothly and steadily leading the team to join Megvii (also known as Face++). Built the beauty and body shaping capabilities from scratch, initially reaching the top level in the industry. Technology Development: Development of face detection and alignment algorithms. Development of algorithms for gender and age detection of faces. Development of beauty and body shaping algorithms (including skin smoothing, spot removal, face slimming, and eye enlargement). Development of algorithms for wide dynamic range scene detection. Development of methods for automatic filter generation.

Image Quality Optimization Depth Estimation OpenCL acceleration for Stereo Vision

REPORT TO General Manager of R&D 1. Build the ISP Group up to 17 members 2. Make schedules to improve Camera Imaging Quality 3. Design and implement general framework of ISP firmware component compatible with all common surveillance SoC platforms. 4. Auto Focus algorithm for PTZ camera.

REPORT TO Senior Project Manager 1. Lead team (3~4 persons) to develop senior Image & Video processing algorithms for Image Deblur, Digital Video Stabilizer, Stitching, and Image Dehaze. 2. Work for ISP pipeline algorithms R&D, and optimization for FPGA platform, such as BLC, DPC, FPN, CFAI, LSC, Contrast Enhancement, Edge Enhancement, Spatial and Temporal Noise Reduction(STNR), and so on.

REPORT TO Manager of R&D 1. Moving objects tracking algorithm development for intelligent traffic products, with feature vector and pattern matching technologies. 2. OCR using projection and machine learning methods, by moment features extracting and SVM & KNN pattern classification for characters’ recognition.

Digital Image Processing: image segmentation, see the publications. Computer Vision：PCA for face recognition Artifical Intelligence Algorithm: for fiber sensor strain calculation Genetic Algorithm(GA)/Ant Colony Optimization(ACO)/Artificial Neural Networks(ANN): see the publications. Pattern Recognition: Support Vector Machine(SVM)/Machine Learning(ML) For data regressive and forecasting, see the publications.

MATLAB, Digital Circuit，Signals and Systems，Image Processing