train.py

import argparse
import json
import numpy as np
import os
from pathlib import Path
import torch.backends.cudnn as cudnn
import torch.utils.data
from torch.utils.tensorboard import SummaryWriter
import torchvision.transforms as transforms
import torchvision.datasets as datasets

import utils.misc as misc
from utils.misc import NativeScalerWithGradNormCount as NativeScaler
import swin_mae
from utils.engine_pretrain import train_one_epoch
from data import ImageDataset
from glob import glob
from torch.utils.data import DataLoader, Dataset


def get_args_parser():
    parser = argparse.ArgumentParser('MAE pre-training', add_help=False)
    # common parameters
    parser.add_argument('--batch_size', default=64, type=int)
    parser.add_argument('--epochs', default=200, type=int)
    parser.add_argument('--save_freq', default=10, type=int)
    parser.add_argument('--checkpoint_encoder', default='', type=str)
    parser.add_argument('--checkpoint_decoder', default='', type=str)
    parser.add_argument('--data_path', default=r'/home/cv/anomaly_detection/full_data/weed_detection/normal_samples', type=str)  # fill in the dataset path here
    parser.add_argument('--mask_ratio', default=0.75, type=float,
                        help='Masking ratio (percentage of removed patches).')

    # model parameters
    parser.add_argument('--model', default='swin_mae', type=str, metavar='MODEL',
                        help='Name of model to train')
    parser.add_argument('--input_size', default=224, type=int,
                        help='images input size')
    parser.add_argument('--norm_pix_loss', action='store_true',
                        help='Use (per-patch) normalized pixels as targets for computing loss')
    parser.set_defaults(norm_pix_loss=False)

    # optimizer parameters
    parser.add_argument('--accum_iter', default=1, type=int)
    parser.add_argument('--weight_decay', type=float, default=0.05,
                        help='weight decay (default: 0.05)')
    parser.add_argument('--lr', type=float, default=1e-3, metavar='LR',
                        help='learning rate (absolute lr)')
    parser.add_argument('--min_lr', type=float, default=0., metavar='LR',
                        help='lower lr bound for cyclic schedulers that hit 0')
    parser.add_argument('--warmup_epochs', type=int, default=10, metavar='N',
                        help='epochs to warmup LR')

    # other parameters
    parser.add_argument('--output_dir', default='./output_pk',
                        help='path where to save, empty for no saving')
    parser.add_argument('--log_dir', default='./output_indices',
                        help='path where to tensorboard log')
    parser.add_argument('--device', default='cuda',
                        help='device to use for training / testing')
    parser.add_argument('--seed', default=0, type=int)
    parser.add_argument('--start_epoch', default=0, type=int, metavar='N',
                        help='start epoch')
    parser.add_argument('--num_workers', default=4, type=int)
    parser.add_argument('--pin_mem', action='store_true',
                        help='Pin CPU memory in DataLoader for more efficient (sometimes) transfer to GPU.')
    parser.set_defaults(pin_mem=True)

    return parser


def main(args):
    # Fixed random seeds
    seed = args.seed
    torch.manual_seed(seed)
    np.random.seed(seed)

    # Set up training equipment
    device = torch.device(args.device)
    cudnn.benchmark = True


    data_folder = "/home/prateekjha/anomaly_detection/full_data/Agriculture-Vision-2021/train/images/rgb_nir"
    batch_size = 64
    device = 'cuda' if torch.cuda.is_available() else 'cpu'
    
    class_name='planter_skip'
    with open("/home/prateekjha/anomaly_detection/full_data/label_dict_train.json") as F:
        label_train_id=json.load(F)
    image_list=glob("/home/prateekjha/anomaly_detection/full_data/Agriculture-Vision-2021/train/images/rgb_nir/*")

    image_paths=[]
    for x in image_list:
        image_id=x.split("/")[-1]
        if image_id not in label_train_id[class_name]:
            image_paths.append(x)


    #all_models=glob("/home/cv/anomaly_detection/MAE/exp_1/*")
    #all_models=sorted(all_models,key=lambda x:int(x.split("/")[-1].split("_")[-1].split(".")[0]))

    #image_paths=glob(f"{data_folder}/*")
    #mask_paths=glob(f"{data_folder}/mask/*")

    image_paths=sorted(image_list,key=lambda x:x.split("/")[-1].split(".")[0])
    #mask_paths=sorted(mask_paths,key=lambda x:x.split("/")[-1].split(".")[0])


    transform_train = transforms.Compose([
        transforms.Resize((224,224)),
        #transforms.RandomHorizontalFlip(),
        transforms.ToTensor()
    ])

    # Create the data loaders for train and test sets
    data_iterator = ImageDataset(image_paths,transform_train)
    data_loader = DataLoader(data_iterator, batch_size=batch_size, shuffle=False, num_workers=4)

    # Log output
    if args.log_dir is not None:
        os.makedirs(args.log_dir, exist_ok=True)
        log_writer = SummaryWriter()
    else:
        log_writer = None

    # Set model
    model = swin_mae.__dict__[args.model](norm_pix_loss=args.norm_pix_loss, mask_ratio=args.mask_ratio)
    model.to(device)
    model_without_ddp = model

    # Set optimizer
    param_groups = [p for p in model_without_ddp.parameters() if p.requires_grad]
    optimizer = torch.optim.AdamW(param_groups, lr=args.lr, weight_decay=5e-2, betas=(0.9, 0.95))  # 原来是5E-2
    loss_scaler = NativeScaler()

    # Create model
    misc.load_model(args=args, model_without_ddp=model_without_ddp)

    args.epochs=100
    # Start the training process
    print(f"Start training for {args.epochs} epochs")
    for epoch in range(args.start_epoch, args.epochs):
        train_stats = train_one_epoch(
            model, data_loader,
            optimizer, device, epoch, loss_scaler,
            log_writer=log_writer,
            args=args
        )
        if args.output_dir and ((epoch + 1) % args.save_freq == 0 or epoch + 1 == args.epochs):
            misc.save_model(
                args=args, model=model, model_without_ddp=model_without_ddp, optimizer=optimizer,
                loss_scaler=loss_scaler, epoch=epoch + 1)

        log_stats = {**{f'train_{k}': v for k, v in train_stats.items()},
                     'epoch': epoch, }

        if args.output_dir and misc.is_main_process():
            if log_writer is not None:
                log_writer.flush()
            with open(os.path.join(args.output_dir, "log.txt"), mode="a", encoding="utf-8") as f:
                f.write(json.dumps(log_stats) + "\n")

if __name__ == '__main__':
    arg = get_args_parser()
    arg = arg.parse_args()
    if arg.output_dir:
        Path(arg.output_dir).mkdir(parents=True, exist_ok=True)
    main(arg)