config.py

import torch
from pathlib import Path

IMG_SIZES = [4, 8, 16, 32, 64, 128, 256, 512, 1024]

### Data
DATA_DIR = "/home/ubuntu/project/cv/celebahq"

### Dataloader
N_WORKERS = 4
AMP = False
# IMG_SIZE_BATCH_SIZE = {4: 16, 8: 16, 16: 16, 32: 16, 64: 16, 128: 16, 256: 14, 512: 6, 1024: 3} # In the paper
IMG_SIZE_BATCH_SIZE = {4: 16, 8: 16, 16: 16, 32: 16, 64: 16, 128: 9, 256: 3, 512: 3, 1024: 3} # In my case

### Loss
LAMBDA = 10 # DO NOT MODIFY
LOSS_EPS = 0.001 # DO NOT MODIFY

### Adam optimizer
LR = 0.001 # DO NOT MODIFY
BETA1 = 0 # DO NOT MODIFY
BETA2 = 0.99 # DO NOT MODIFY
ADAM_EPS = 1e-8 # DO NOT MODIFY

### Training
N_GPUS = torch.cuda.device_count()
MULTI_GPU = False
N_PRINT_STEPS = 1000
N_VAL_STEPS = 1000
# "We start with 4×4 resolution and train the networks until we have shown the discriminator
# 800k real images in total. We then alternate between two phases: fade in the first 3-layer block
# during the next 800k images, stabilize the networks for 800k images, fade in the next 3-layer block
# during 800k images, etc."
IMG_SIZE_N_IMAGES = {
    4: 200_000,
    8: 200_000,
    16: 400_000,
    32: 400_000,
    64: 800_000,
    128: 800_000,
    256: 800_000,
    512: 800_000,
    1024: 800_000,
}

### Checkpoint
CKPT_PATH = Path(__file__).parent/"checkpoints/512×512_736000.pth"
STEP = None
TRANS_PHASE = None
IMG_SIZE_IDX = None

### Pre-trained Model
PRETRAINED = "/home/ubuntu/project/cv/pggan_from_scratch/pretrained/pggan_512.pth"