
GAN1 A to Z code making: https://github.com/KiHwanLee123/Coursera---GAN/blob/main/GAN1.ipynb판별자와 생성자판별자 : P(Class|data)생성자 : 노이즈와 class를 받아 특정 fe

Deep Convolutional GAN (DCGAN) A to Z code making: https://github.com/KiHwanLee123/Coursera---GAN/blob/main/Deep%20Convolutional%20GAN%20(DCGAN).

위와 같은 경우에는 Gradient를 계산하여 파라미터 업데이트를 잘 수행하게 된다. 하지만 이 경우와 같이 분포가 꽤나 떨어져 있다면, Vanishing Gradient문제로 초기 업데이트가 잘 되지 않게 된다. 이것이 Binary Cross Entropy Loss의

이미지는 빨강, 녹색 또는 파랑(RGB) 색상의 강도를 나타내는 0에서 256까지의 정수로 표현됩니다. 단백질도 마찬가지로 문자를 사용하여 아래와 같이 20개의 고유한 아미노산을 나타냅니다: >MKYATLLEYAFQALKNSYAPYSRFRVGAALLSDDGEVVTGC

GAN의 평가가 어려운 이유random noise를 통해 genenrate했는데, 어떻게 평가할 것인가?마치 명작을 그리는 법을 배우는 고급 예술가와 같다.옳고 그름이 명확히 없다.그렇다면 어떤 관점으로 평가하는 것이 좋을까?1) Fidelity생성 된 이미지의 품질,

GAN의 단점으로는 단방향이라는 점이다.즉, 이미지를 latent vector로 정확하게 보낼 수 없다는 것인데 왜 이것이 단점이 될까?우리는 단순히 이미지를 generate하는 것이 목적이 아닌가?=> 이에 대한 이유로는 이미지 editing을 실시할 수 있어지기 때

Progressive growing, Noise mapping network, Adaptive instance normalization(AdaIN)GAN Improvements1) Stability - Mode collapse미니 배치에서 편향이 있으면 이렇게 GAN에

Image to Image1) GauGAN2) Super Resolution – sharpened image3) multimodal image to image translation ex) cat to dog4) Text to image5) Image and Landma

Image to Image1) GauGAN2) Super Resolution – sharpened image3) multimodal image to image translation ex) cat to dog4) Text to image5) Image and Landma

Image to Image1) GauGAN2) Super Resolution – sharpened image3) multimodal image to image translation ex) cat to dog4) Text to image5) Image and Landma

Image to Image1) GauGAN2) Super Resolution – sharpened image3) multimodal image to image translation ex) cat to dog4) Text to image5) Image and Landma

Image to Image1) GauGAN2) Super Resolution – sharpened image3) multimodal image to image translation ex) cat to dog4) Text to image5) Image and Landma

Image to Image1) GauGAN2) Super Resolution – sharpened image3) multimodal image to image translation ex) cat to dog4) Text to image5) Image and Landma

Image to Image1) GauGAN2) Super Resolution – sharpened image3) multimodal image to image translation ex) cat to dog4) Text to image5) Image and Landma

흑백에서 컬러 이미지로 변환, segmentation map에서 실제 이미지로 변환, video to video, daty to night, edges to photopaired image to imge translation : 지정해준 pose를 수행하게 끔 생성 이

이제는 두 이미지 간을 매핑시켜 차이점을 찾고, 적용시키는 방안을 진행합니다.예를 들어 paired generation은 아래와 같이 edge를 따서 생성할 수 있습니다.그에 반해 unpaired generation은 말을 얼룩말로 바꾸거나, 아래의 모네 그림을 실제로