run/run_configs.py

import os
import argparse
from mmengine import Config

def create_deeplabv3plus_config(model_config_path, dataset_config_path, num_class, work_dir, save_dir, batch_size, max_iters, val_interval):
    cfg = Config.fromfile(model_config_path)
    dataset_cfg = Config.fromfile(dataset_config_path)
    cfg.merge_from_dict(dataset_cfg)
    
    # Set crop size
    cfg.crop_size = (512, 512)
    cfg.model.data_preprocessor.size = cfg.crop_size

      # Configure normalization
    cfg.norm_cfg = dict(type='BN', requires_grad=True)
    cfg.model.backbone.norm_cfg = cfg.norm_cfg
    cfg.model.decode_head.norm_cfg = cfg.norm_cfg
    cfg.model.auxiliary_head.norm_cfg = cfg.norm_cfg

    cfg.model.decode_head.num_classes = num_class
    cfg.model.auxiliary_head.num_classes = num_class

    cfg.train_dataloader.batch_size = batch_size
    
    # Set training configurations
    cfg.train_cfg.max_iters = max_iters
    cfg.train_cfg.val_interval = val_interval
    cfg.default_hooks.logger.interval = 100
    cfg.default_hooks.checkpoint.interval = 2500
    cfg.default_hooks.checkpoint.max_keep_ckpts = 1
    cfg.default_hooks.checkpoint.save_best = 'mIoU'

    cfg['randomness'] = dict(seed=0)
    # Set work directory
    cfg.save_dir = save_dir
    name = os.path.basename(dataset_config_path).split('_')[0] + "_" + os.path.dirname(model_config_path).split(os.sep)[1]
    cfg.work_dir = os.path.join(work_dir,name)
    os.makedirs(cfg.work_dir, exist_ok=True)
    save_config_file = os.path.join(save_dir, f"{name}.py")
    cfg.dump(save_config_file)
    print(f"Configuration saved to: {save_config_file}")

def create_knet_config(model_config_path, dataset_config_path, num_class, work_dir, save_dir, batch_size, max_iters, val_interval):
   
    cfg = Config.fromfile(model_config_path)
    dataset_cfg = Config.fromfile(dataset_config_path)
    
    cfg.merge_from_dict(dataset_cfg)
    
    cfg.norm_cfg = dict(type='BN', requires_grad=True)
    cfg.model.data_preprocessor.size = cfg.crop_size

    cfg.model.decode_head.kernel_generate_head.num_classes = num_class
    cfg.model.auxiliary_head.num_classes = num_class

    cfg.train_dataloader.batch_size = batch_size
    cfg.work_dir = work_dir

    cfg.train_cfg.max_iters = max_iters
    cfg.train_cfg.val_interval = val_interval
    cfg.default_hooks.logger.interval = 100
    cfg.default_hooks.checkpoint.interval = 2500
    cfg.default_hooks.checkpoint.max_keep_ckpts = 1
    cfg.default_hooks.checkpoint.save_best = 'mIoU'

    cfg['randomness'] = dict(seed=0)

    cfg.save_dir = save_dir
    name = os.path.basename(dataset_config_path).split('_')[0] + "_" + os.path.dirname(model_config_path).split(os.sep)[1]
    cfg.work_dir = os.path.join(work_dir, name)
    os.makedirs(cfg.work_dir, exist_ok=True)
    save_config_file = os.path.join(save_dir, f"{name}.py")
    cfg.dump(save_config_file)
    print(f"Configuration saved to: {save_config_file}")

def create_mask2former_config(model_config_path, dataset_config_path, num_class, work_dir, save_dir, batch_size, max_iters, val_interval):
    cfg = Config.fromfile(model_config_path)
    dataset_cfg = Config.fromfile(dataset_config_path)
    cfg.merge_from_dict(dataset_cfg)
    
    # Set crop size
    cfg.crop_size = (512, 512)
    cfg.model.data_preprocessor.size = cfg.crop_size

      # Configure normalization
    cfg.norm_cfg = dict(type='BN', requires_grad=True)

    cfg.model.decode_head.num_classes = num_class
    cfg.model.decode_head.loss_cls.class_weight = [1.0] * num_class + [0.1]

    cfg.train_dataloader.batch_size = batch_size
    
    # Set training configurations
    cfg.train_cfg.max_iters = max_iters
    cfg.train_cfg.val_interval = val_interval
    cfg.default_hooks.logger.interval = 100
    cfg.default_hooks.checkpoint.interval = 2500
    cfg.default_hooks.checkpoint.max_keep_ckpts = 1
    cfg.default_hooks.checkpoint.save_best = 'mIoU'

    cfg['randomness'] = dict(seed=0)
    # Set work directory
    cfg.save_dir = save_dir
    name = os.path.basename(dataset_config_path).split('_')[0] + "_" + os.path.dirname(model_config_path).split(os.sep)[1]
    cfg.work_dir = os.path.join(work_dir,name)
    os.makedirs(cfg.work_dir, exist_ok=True)
    save_config_file = os.path.join(save_dir, f"{name}.py")
    cfg.dump(save_config_file)
    print(f"Configuration saved to: {save_config_file}")

def create_segformer_config(model_config_path, dataset_config_path, num_class, work_dir, save_dir, batch_size, max_iters, val_interval):
    cfg = Config.fromfile(model_config_path)
    dataset_cfg = Config.fromfile(dataset_config_path)
    cfg.merge_from_dict(dataset_cfg)

      # Configure normalization
    cfg.norm_cfg = dict(type='BN', requires_grad=True)
    cfg.model.data_preprocessor.size = cfg.crop_size
    cfg.model.decode_head.norm_cfg = cfg.norm_cfg

    cfg.model.decode_head.num_classes = num_class

    cfg.train_dataloader.batch_size = batch_size
    
    # Set training configurations
    cfg.train_cfg.max_iters = max_iters
    cfg.train_cfg.val_interval = val_interval
    cfg.default_hooks.logger.interval = 100
    cfg.default_hooks.checkpoint.interval = 2500
    cfg.default_hooks.checkpoint.max_keep_ckpts = 1
    cfg.default_hooks.checkpoint.save_best = 'mIoU'

    cfg['randomness'] = dict(seed=0)
    # Set work directory
    cfg.save_dir = save_dir
    name = os.path.basename(dataset_config_path).split('_')[0] + "_" + os.path.dirname(model_config_path).split(os.sep)[1]
    cfg.work_dir = os.path.join(work_dir,name)
    os.makedirs(cfg.work_dir, exist_ok=True)
    save_config_file = os.path.join(save_dir, f"{name}.py")
    cfg.dump(save_config_file)
    print(f"Configuration saved to: {save_config_file}")

def main():
    parser = argparse.ArgumentParser(description='Train configuration setup for different models.')
    
    parser.add_argument('--model_name', type=str, required=True, choices=['deeplabv3plus', 'knet', 'mask2former', 'segformer'],
                        help='Model name to generate the config for.')
    parser.add_argument('-m', '--model_config', type=str, required=True, help="Path to the model config file")
    parser.add_argument('-d', '--dataset_config', type=str, required=True, help='Path to the dataset config file.')
    parser.add_argument('-c', '--num_class', type=int, required=True, help="Number of classes in the dataset")
    parser.add_argument('-w','--work_dir', type=str, required=True, help='Directory to save the train result.')
    parser.add_argument('-s', '--save_dir', type=str, required=True, help="Directory to save the generated config file")
    parser.add_argument('--batch_size', type=int, default=2, help='Batch size for training.')
    parser.add_argument('--max_iters', type=int, default=20000, help='Number of training iterations.')
    parser.add_argument('--val_interval', type=int, default=500, help='Interval for validation during training.')


    args = parser.parse_args()

    if args.model_name == 'deeplabv3plus':
        create_deeplabv3plus_config(
            model_config_path=args.model_config,
            dataset_config_path=args.dataset_config,
            num_class=args.num_class,
            work_dir=args.work_dir,
            save_dir =args.save_dir,
            batch_size=args.batch_size,
            max_iters=args.max_iters,
            val_interval=args.val_interval
        )
    if args.model_name == 'knet':
        create_knet_config(
            model_config_path=args.model_config,
            dataset_config_path=args.dataset_config,
            num_class=args.num_class,
            work_dir=args.work_dir,
            save_dir =args.save_dir,
            batch_size=args.batch_size,
            max_iters=args.max_iters,
            val_interval=args.val_interval
        )
    if args.model_name == 'mask2former':
            create_mask2former_config(
            model_config_path=args.model_config,
            dataset_config_path=args.dataset_config,
            num_class=args.num_class,
            work_dir=args.work_dir,
            save_dir =args.save_dir,
            batch_size=args.batch_size,
            max_iters=args.max_iters,
            val_interval=args.val_interval
        )
    elif args.model_name == 'segformer':
            create_segformer_config(
            model_config_path=args.model_config,
            dataset_config_path=args.dataset_config,
            num_class=args.num_class,
            work_dir=args.work_dir,
            save_dir =args.save_dir,
            batch_size=args.batch_size,
            max_iters=args.max_iters,
            val_interval=args.val_interval
        )

if __name__ == '__main__':
    main()