doublejy715 / Paper_review

Paper_review
1 stars 0 forks source link

StyleMapGAN : Exploiting Spatial Dimensions of Latent in GAN for Real-time Image Editing #18

Open doublejy715 opened 2 years ago

doublejy715 commented 2 years ago

Abstract

1. Introduction

paragraph 1

GAN의 설명과 최근 도전과제

paragraph 2

image-to-image translation을 위한 2가지 방법

paragraph 3

image-to-image translation을 위한 2가지 방법

paragraph 4

StyleMapGAN설명 및 특징

2. Related work

Optimization-based editing methods

Learning-based editing methods

Local editing methods

3. StyleMapGAN

3.1. Stylemap-based generator

paragraph 1

Stylemap의 간단한 구조 설명 image

paragraph 2

Synthesis network에 stylemap이 적용되는 식

3.2. Training procedure and losses

image

3.3. Local editing

paragraph 1

local editing을 수행하는 방법들

paragraph 2

mask를 이용한 새로운 latent vector :w 의 소개

paragraph 3

8x8 size의 mask 부터 사용하는 이유

4. Experiments

4.1. Experimental setup

Baselines

Datasets

4.2. Evaluation metrics

Frechet inception distance(FID)

FID_{lerp}

MSE & LPIPS

Average precision(AP)

MSE{src} & MSE{ref}

4.3. Effects of stylemap resolution

paragraph 1

spatial resolution의 이점

paragraph 2

paragraph 3

4.4. Real image projection

4.5. Local editing

4.6. Unaligned tranplantation

5. Discussion and Conclusion