The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) 3D Deep Learning with PyTorch 3D w/ Georgia Gkioxari - #408
Sep 10, 2020
Georgia Gkioxari, a research scientist at Facebook AI Research, shares her insights on the groundbreaking open-source library PyTorch3D. She discusses her journey from traditional object recognition to deep learning innovations and the evolution of 3D understanding. Georgia highlights the user experience and modularity of PyTorch3D, revealing its role in enhancing machine learning capabilities. She also reflects on her responsibilities as co-chair for CVPR 2021 and the need for modernizing academic peer review processes to adapt to growing challenges in research.
AI Snips
Chapters
Transcript
Episode notes
Deep Learning's Impact
- Deep learning revolutionized computer vision after 2012.
- The shift towards modular, end-to-end models and open-source tools accelerated progress.
PyTorch 3D's Purpose
- PyTorch 3D enables efficient 3D data computations, addressing limitations of 2D-focused deep learning models.
- It supports various 3D data types and differentiable rendering, expanding possibilities in computer vision.
PyTorch 3D's Flexibility
- PyTorch 3D handles both explicit 3D data from sensors and latent 3D representations within neural networks.
- Its differentiable operators allow flexible integration into deep learning models.