The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) Trends in Computer Vision with Pavan Turaga - #444
Jan 4, 2021
Pavan Turaga, an Associate Professor from Arizona State University, dives into the latest trends in computer vision. He discusses the exciting revival of physics-based scene analysis and the evolution of differentiable rendering, emphasizing its role in 3D structure reconstruction. Turaga highlights the significance of self-supervised learning techniques and innovative network architectures that enhance model performance. He also tackles the real-world evaluation challenges for AI systems, offering insights into assessing model reliability and robustness in practical applications.
AI Snips
Chapters
Transcript
Episode notes
Computer Vision Trends 2020
- Computer vision trends in 2020 include revisiting physics-based scene understanding and merging it with neural networks.
- Established trends like self-supervised learning are also finding more applications.
Differentiable Rendering
- Differentiable rendering allows learning implicit 3D representations from 2D images without 3D ground truth.
- This involves rendering scenes from implicit representations and comparing them to input images.
Neural Radiance Fields (NERF)
- NERF (Neural Radiance Fields) represents scenes as light rays, learning light fields implicitly from 2D images.
- The key is differentiable rendering of light fields, enabling end-to-end training without 3D or light field ground truth.