AMontiB
Add utilities and CLI for PGC pipeline
c55595d
Raw History Blame Contribute Delete
3.8 kB
import logging
import os
from typing import Tuple
def setup_logging(
checkpoints_dir: str,
exp_name: str,
level: int = logging.INFO,
log_type: str = "train",
) -> Tuple[logging.Logger, str]:
os.makedirs(checkpoints_dir, exist_ok=True)
if log_type == "test":
log_file = os.path.join(checkpoints_dir, f"{exp_name}_test.log")
else:
log_file = os.path.join(checkpoints_dir, f"{exp_name}.log")
root_logger = logging.getLogger()
root_logger.setLevel(level)
if root_logger.handlers:
root_logger.handlers.clear()
formatter = logging.Formatter(
"%(asctime)s - %(name)s - %(levelname)s - %(message)s"
)
file_handler = logging.FileHandler(log_file)
file_handler.setLevel(level)
file_handler.setFormatter(formatter)
stream_handler = logging.StreamHandler()
stream_handler.setLevel(level)
stream_handler.setFormatter(formatter)
root_logger.addHandler(file_handler)
root_logger.addHandler(stream_handler)
return root_logger, log_file
def log_training_config(logger: logging.Logger, opt) -> None:
"""Log the training configuration in a structured, human-readable form.
All field accesses are direct: every attribute below is guaranteed to be
populated by :func:`utils.cli.build_train_parser`.
"""
logger.info("==== Training configuration ====")
logger.info("Name: %s", opt.name)
logger.info("Checkpoints dir: %s", opt.checkpoints_dir)
logger.info("Devices: %s", opt.devices)
logger.info("Batch size: %d", opt.batch_size)
logger.info("Learning rate: %.6f", opt.lr)
logger.info("Optimizer: %s", opt.optim)
logger.info("Weight decay: %.4f", opt.weight_decay)
logger.info("Label smoothing: %.2f", opt.label_smoothing)
logger.info("Number of epochs: %d", opt.niter)
logger.info("Crop size: %d", opt.cropSize)
logger.info("DINO variant: %s", opt.dino_variant)
logger.info("LoRA rank: %d", opt.lora_rank)
logger.info("LoRA alpha: %.2f", opt.lora_alpha)
logger.info("LoRA dropout: %.2f", opt.lora_dropout)
logger.info("Real image dirs (%d):", len(opt.real_image_dir))
for idx, real_dir in enumerate(opt.real_image_dir):
logger.info(" [%d] %s", idx, real_dir)
logger.info("Fake image dirs (%d):", len(opt.fake_image_dir))
for idx, fake_dir in enumerate(opt.fake_image_dir):
logger.info(" [%d] %s", idx, fake_dir)
logger.info("Test root: %s", opt.test_root)
logger.info("DINO pretrained root: %s", opt.dino_pretrained_root)
logger.info("PGCM tau (RGB): %.4f", opt.tau_rgb)
logger.info("PGCM tau (Residual): %.4f", opt.tau_res)
logger.info("Accumulation steps: %d", opt.accumulation_steps)
logger.info("Max grad norm: %.4f", opt.max_grad_norm)
logger.info("Seed: %d", opt.seed)
def log_test_config(logger: logging.Logger, args) -> None:
"""Log the test configuration in a structured, human-readable form.
All attribute accesses are direct (guaranteed by
:func:`utils.cli.build_test_parser`).
"""
logger.info("==== Test configuration ====")
logger.info("Checkpoint: %s", args.checkpoint)
logger.info("Test root: %s", args.test_root)
logger.info("Batch size: %d", args.batch_size)
logger.info("Num workers: %d", args.num_workers)
logger.info("Devices: %s", args.devices)
logger.info("Crop size: %d", args.cropSize)
logger.info("DINO variant: %s", args.dino_variant)
logger.info("LoRA rank: %d", args.lora_rank)
logger.info("LoRA alpha: %.2f", args.lora_alpha)
logger.info("LoRA dropout: %.2f", args.lora_dropout)
logger.info("DINO pretrained root: %s", args.dino_pretrained_root)
logger.info("PGCM tau (RGB): %.4f", args.tau_rgb)
logger.info("PGCM tau (Residual): %.4f", args.tau_res)