[go: up one dir, main page]

Skip to content
/ PGGAN Public

PyTorch implementation of 'PGGAN' (Karras et al., 2018) from scratch and training it on CelebA-HQ at 512 × 512

Notifications You must be signed in to change notification settings

KimRass/PGGAN

Repository files navigation

1. Pre-trained Model

2. Samples

1) Drawn During Training

4 × 4
8 × 8
16 × 16
32 × 32
64 × 64
128 × 128
256 × 256
512 × 512
1,024 × 1,024

2) Drawn From the Trained Model

3. Implementation Details

1) Normalization

  • 이전까지는 CelebA-HQ의 Training set에 대해 Mean과 Std를 계산해서 이걸 가지고 Normalize했습니다.
    • T.Normalize(mean=(0.517, 0.416, 0.363), std=(0.303, 0.275, 0.269))
  • 논문에서는 "We represent training and generated images in $[-1, 1]$."라는 표현이 있는 것을 알고 있었으나 이를 간과하고 있었습니다.
  • 다음과 같이 코드를 수정했습니다.
    • T.Normalize(mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5))

About

PyTorch implementation of 'PGGAN' (Karras et al., 2018) from scratch and training it on CelebA-HQ at 512 × 512

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published