Per Garment Capture and Synthesis for Real-time Virtual Try-on
summary
A pipeline captures each garment with a 9-DOF actuated mannequin and a color-coded measurement garment to record body shape and pose, collecting about 80,000 images per garment in roughly two hours, then trains a pix2pixHD-based image-to-image translation for ~2 days on paired measurement/target frames. At runtime the system segments the measurement shirt, maps to the target garment, swaps weights live and composites results in real time. Evaluation shows practical size prediction—7 of 8 users pick the correct size.
AI-written summary · about
related talks
33mSee More, Be More
46mImage Recognition with Deep Learning From prototype to production
15mMeCap: Whole-Body Digitization for Low-Cost VR/AR Headsets
41mComputer Vision for Remote AR
16mEye-Trace: Segmentation of Volumetric Microscopy Images with Eyegaze
17mFrom recovering Pro Golfer to AI Entrepreneur
35mDeep learning for 3d human motion capture
20mInteractive Room Capture on 3D-Aware Mobile Devices