train.py

import os
from datetime import datetime
import argparse

from lib.utils import yaml2config
from networks.model32 import AdversarialModel as AdversarialModel32
from networks.model64 import AdversarialModel as AdversarialModel64
import torch

try:
    from torch.utils.tensorboard import SummaryWriter
except ModuleNotFoundError:
    from tensorboardX import SummaryWriter


if __name__ == "__main__":
    parser = argparse.ArgumentParser(description="config")
    parser.add_argument(
        "--config",
        nargs="?",
        type=str,
        default="./configs/SpiS_gan_iam_32.yml",
        help="Configuration file to use",
    )

    args = parser.parse_args()
    print(f"Config file: {args.config}")

    cfg = yaml2config(args.config)
    run_id = datetime.strftime(datetime.now(), '%m-%d-%H-%M')
    logdir = os.path.join("runs", os.path.basename(args.config)[:-4] + '-' + str(run_id))
    print(logdir)
    
    DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    if DEVICE.type == "cuda":
        print(f"[INFO] CUDA available. Using device: {DEVICE} - {torch.cuda.get_device_name(DEVICE)}")
    else:
        print("[INFO] CUDA not available. Falling back to CPU.")

    cfg['device'] = str(DEVICE)

    if(cfg.img_height==64):
        all_models = {
            'adversarial_model': AdversarialModel64
        }
    else:
        all_models = {
            'adversarial_model': AdversarialModel32
        }

    def get_model(name):
     return all_models[name]
    
    model = get_model(cfg.model)(cfg, logdir)
    
    # Check and load checkpoint
    epoch_done = 1
    if cfg.ckpt and os.path.exists(cfg.ckpt):
        print(f"Loading checkpoint from {cfg.ckpt}")
        epoch_done = model.load(cfg.ckpt, cfg.device)
    else:
        print("No valid checkpoint found, starting from scratch.")

    model.train(epoch_done=epoch_done)