Computer Vision in the Era of Deep Learning
summary
Explains how convolutional filters become learned features in CNNs and why Viola‑Jones cascades are now replaced by learned kernels; shows a concrete monocular depth pipeline (encoder‑decoder trained with stereo pairs, credited to godart 2017) that produces per‑pixel disparity from a single rear‑mirror camera; details a blind face‑restoration recipe using a codebook + Transformer (joan 2022) to convert low‑quality or synthetic faces into realistic images; reports a real use case where restored synthetic faces boost gender‑estimation accuracy by about 5%.
AI-written summary · about
chapters (9)
- 0:00Introduction to AI and vision
- 2:07Computer vision basics and challenges
- 7:06Edge detection and filters
- 9:36Neural networks in computer vision
- 12:50Deep learning breakthroughs
- 14:41Monocular depth estimation
- 19:00Face restoration with Transformers
- 23:16Synthetic data and bias management
- 24:42Conclusion and Q&A
related talks
47mDeep Learning in Computer Vision
01h 18mComputer Vision after the Victory of Data
04mUsing Computer Vision to Understand Big Visual Data
14mA Primer to Modern Day Computer Vision
42mState of the Art of Innovation in Computer Vision
42mDeep Learning for Computer Vision
54mIntroduction to Understanding Deep Learning
01h 30mDeep Learning Techniques for Monocular Depth Estimation