The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) Optical Flow Estimation, Panoptic Segmentation, and Vision Transformers with Fatih Porikli - #579
20 snips
Jun 20, 2022 
Fatih Porikli, Senior Director of Engineering at Qualcomm AI Research, discusses groundbreaking advancements in computer vision. Topics include a cutting-edge framework for panoptic segmentation that combines semantic and instance contexts, and novel strategies for optical flow estimation enhancing accuracy. He also delves into the IRISformer, a transformer model designed for rendering complex indoor scenes from single images. Additionally, Fatih highlights the importance of workshops and practical demos at the CVPR conference to engage and inspire future innovations.
AI Snips
Chapters
Transcript
Episode notes
Bridging Academia and Industry
- Fatih Porikli has transitioned between academia and industry multiple times, driven by a desire to solve increasingly complex problems.
- His focus remains computer vision, reflecting the significant role visual perception plays in human brain activity.
Expanding Computer Vision
- Computer vision extends beyond images and videos to encompass 3D data and radio frequency signals.
- These modalities enable applications like augmented reality, autonomous vehicles, and robotics, tackling unsolved challenges.
Understanding Panoptic Segmentation
- Panoptic segmentation labels every pixel with an identity, differentiating between countable "things" and uncountable "stuff".
- This complex task combines instance segmentation (identifying things) and semantic segmentation (identifying stuff).