C-Face: Continuously reconstructing facial expressions by deep learning contours of the face ...
summary
Ear-mounted RGB cameras embedded into headphones and earphones capture facial contour silhouettes and a CNN regresses 42 landmarks (mouth, eyes, eyebrows) producing continuous reconstructions with a reported 1.39–1.43 mm MMAE across nine participants; comparisons use dlib from a front-facing camera as ground truth. System handles users wearing masks, benefits from data augmentation for remounting, is currently user-dependent, and fails when hair occludes cameras. ear-mounted RGB cameras
AI-written summary · about
chapters (17)
- 0:00Intro
- 0:24Motivation
- 0:59Most Popular Approach
- 1:18Problem
- 2:26Obvious Solution ..........
- 2:45Previous Research
- 4:09C-Face Solution
- 5:30Preliminary Study
- 6:23Deep learning
- 6:39Ground-truth Collection
- 6:53Evaluation Metric
- 8:05Data Collection
- 8:41Continuous Reconstruction Study
- 10:33Remounting Study - Device shifting
- 11:55Face Blocking Study
- 12:40Limitation and Future work
- 14:13Conclusion
related talks
06mAcoustic-based Upper Facial Action Recognition for Smart Eyewear
38m3D Face Tracking and Reconstruction using Modern C++
52mFace Recognition: From Scratch To Hatch
23mMultimodal Video Face Liveness: A Better Alternative?
32mHeart to Heart and Eye to Eye with Deep Fakes
57mFoundations of Facial Recognition
40m"Introduction to Face Processing with Computer Vision|"
31mFace Recognition