<style> .reveal, .reveal h1, .reveal h2, .reveal h3, .reveal h4, .reveal h5, .reveal h6 { font-family: "Kai"} .rightalign {text-align:right} </style> # <h2>Yes, we GAN: Applying Adversarial Techniques for Autonomous Driving</h2> ##### 報告人: 陳雪芬 ##### 報告人: 陳居億 ##### 時間: 2020/10/15 ###### https://hackmd.io/@NtutShare/SJ34R9EPw --- ## Outline 1. Introduction 2. Overview of GAN 3. GAN Applications for autonomous driving 4. Our Results 5. Discussion 6. Conclusions <!-- 2. Problem Formulation/Research Question/Theoretical Framework 2. Challenges, Issues & Difficulties 2. Methodology/Case Study 3. Experimental Results/Numerical Testing/Analysis 4. Discussion/Conclusion --> --- ## Introduction  --- ## Overview of GAN - Vanilla GAN - Prominent Derivatives of GAN - GAN:Recent Advances --- ### Vanilla GAN  --- ### Vanilla GAN - Advantage - back-propagation - wide variety of functions - Disadvantage - explicit representation of $p_g(x)$ - simultaneous optimization --- ### Prominent Derivatives of GAN - e.g. - Conditional Generative Adversarial Nets (CGAN) - Wasserstein GAN (WGAN) - Improved WGAN - Boundary-Seeking Generative Adversarial Networks(BGAN) --- ### GAN:Recent Advances - e.g. - BigGAN - Self-Attention Generative Adversarial Network (SAGAN) --- ## GAN Applications for autonomous driving - Advanced Data Augmentation <!-- - Semi-supervised/Unsupervised Learning - Learned Loss Functions --> - Adversarial training/testing --- ### Advanced Data Augmentation  --- ### Advanced Data Augmentation - GAN create realitic looking images - e.g. - CGAN、CycleGAN+UNIT、AC-GAN --- ### Advanced Data Augmentation - 2D Synthesis - Pix2Pix、SRGAN、CycleGAN、DiscoGAN、StarGAN、UNIT、Pix2PixHD、BicycleGAN、MUNIT、Augmented GAN  --- ### Advanced Data Augmentation - 2D Synthesis  --- ### Advanced Data Augmentation - 3D Synthesis - 3D-GAN、PrGAN、PC-GAN  --- ### Advanced Data Augmentation - Video Synthesis - TGAN - <iframe width="560" height="315" src="https://www.youtube.com/embed/dkoi7sZvWiU" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe> --- ### Advanced Data Augmentation - Object Detection - SeGAN、Perceptual-GAN <!--  !-->  --- ### Advanced Data Augmentation - Domain Adaptation - Pixel level、GraspGAN  --- ### Advanced Data Augmentation - Inpainting  --- ### Advanced Data Augmentation - Super Resolution - SRGAN  --- ### Advanced Data Augmentation - Super Resolution <iframe width="560" height="315" src="https://www.youtube.com/embed/z-ZJqd4eQrc" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe> --- ### Adversarial training/testing - EL-GAN -  --- ## Our Results  --- ## Our Results - e.g. - CycleGAN、MUNIT  --- ## Discussion - Quantitative Evaluation - Adversarial examples and Safety - Optimization Stability --- ## Conclusions - autonomous driving applications - main challenges and open problems --- ## 謝謝聆聽