<style> .reveal, .reveal h1, .reveal h2, .reveal h3, .reveal h4, .reveal h5, .reveal h6 { font-family: "Kai"} .rightalign {text-align:right} </style> # <h2>Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks</h2> ##### 報告人: 陳雪芬 ##### 報告人: 陳居億 ##### 時間: 2020/10/22 ###### https://hackmd.io/@NtutShare/Syfc_qpwv --- ## Outline 1. Introduction 2. Formulation 4. Implementation 11. Limitations and Discussion --- ## Introduction :::info Image-to-image translation <!-- .element: class="fragment" data-fragment-index="1" --> :::  <!--  --> --- ## Introduction #### Unpaired Image-to-Image Translation  --- ## Introduction #### Cycle Consistency  --- ## Introduction #### Neural Style Transfer - mapping between two image collections, rather than between two specific images - capture correspondences between higher-level appearance structures --- ## Formulation  --- ## Formulation #### adversarial losses  --- ## Formulation #### cycle consistency losses  --- ## Formulation #### cycle consistency losses  --- ## Formulation #### Full Objective  :::danger <!-- .element: class="fragment" data-fragment-index="1" --> ::: --- ## Implementation --- #### Network Architecture ###### generative networks  --- #### Network Architecture ##### discriminator networks(from PatchGANs) - PatchGANs - fully convolutional neural networks - more effective - focus on surface-level features  <!--  --> <!-- - a patch-level discriminator architecture has fewer parameters than a full-image discriminator and can work on arbitrarily-sized images in a fully convolutional fashion --> --- #### Results  --- #### Comparation  --- ## Limitations and Discussion #### Limitations - characteristics of the training datasets - geometric changes  --- ## Limitations and Discussion #### Discussion - lingering gap between paired and unpaired --- ## 謝謝聆聽