State of the Art of Innovation in Computer Vision
summary
Survey of CVPR 2018–2019 highlights concrete shifts: the ImageNet/AlexNet jump (20% error improvement) enabled end-to-end convolutional pipelines running on GPUs (early hacks used PlayStation GPUs), GANs evolve into controllable Style-like generators, and joint multi-task nets improve segmentation/pose/depth together. Enterprise moves toward few-shot learning to cut million-image needs. Examples include a patent-stage automated retinal screening application and practical use of OpenPose/DensePose for ranking exercise form rather than exact joint reconstruction.
AI-written summary · about
chapters (8)
- 0:00Introduction and background
- 6:21AI craze and deep learning
- 13:10Compositional learning in CV
- 18:53Advanced architectures overview
- 22:01Theory and applications in CV
- 28:53Visual understanding tasks
- 34:38Personal research approach
- 41:50Closing and Q&A invitation
related talks
14mA Primer to Modern Day Computer Vision
41mDeep Learning to Solve Challenging Problems
04mUsing Computer Vision to Understand Big Visual Data
26mComputer Vision in the Era of Deep Learning
42mDeep Learning for Computer Vision
03mRethinking software engineering in the AI era
01h 18mComputer Vision after the Victory of Data
41mPractical Aspects of Applying Modern Computer Vision Technologies