main_pretrain.py

# --------------------------------------------------------
# References:
# DeiT: https://github.com/facebookresearch/deit
# BEiT: https://github.com/microsoft/unilm/tree/master/beit
# Un-Mix: https://github.com/szq0214/Un-Mix
# MAE: https://github.com/facebookresearch/mae
# --------------------------------------------------------
import argparse
import datetime
import json
import numpy as np
import os
import time
from pathlib import Path

import torch
import torch.backends.cudnn as cudnn
from torch.utils.tensorboard import SummaryWriter
import torchvision.transforms as transforms
import torchvision.datasets as datasets

import wandb
import timm

#assert timm.__version__ == "0.3.2"  # version check
import timm.optim.optim_factory as optim_factory

import util.misc as misc
from util.misc import NativeScalerWithGradNormCount as NativeScaler
import models_mae
#from fb_mae.iMAE_cluster.odataset import classmixINPair

from engine_pretrain import train_one_epoch


def get_args_parser():
    parser = argparse.ArgumentParser('MAE pre-training', add_help=False)
    parser.add_argument('--batch_size', default=64, type=int,
                        help='Batch size per GPU (effective batch size is batch_size * accum_iter * # gpus')
    parser.add_argument('--epochs', default=400, type=int)
    parser.add_argument('--accum_iter', default=1, type=int,
                        help='Accumulate gradient iterations (for increasing the effective batch size under memory constraints)')

    # Model parameters
    parser.add_argument('--model', default='mae_vit_base_patch16_dec512d8b', type=str, metavar='MODEL',
                        help='Name of model to train')

    parser.add_argument('--teach_model', default='tmae_vit_base_patch16_dec512d8b', 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('--mask_ratio', default=0.75, type=float,
                        help='Masking ratio (percentage of removed patches).')

    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('--weight_decay', type=float, default=0.05,
                        help='weight decay (default: 0.05)')

    parser.add_argument('--lr', type=float, default=None, metavar='LR',
                        help='learning rate (absolute lr)')
    parser.add_argument('--blr', type=float, default=1e-3, metavar='LR',
                        help='base learning rate: absolute_lr = base_lr * total_batch_size / 256')
    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=40, metavar='N',
                        help='epochs to warmup LR')

    # Dataset parameters
    parser.add_argument('--data_path', default='/imagenet', type=str,
                        help='dataset path')

    parser.add_argument('--output_dir', default='/output_dir_IN',
                        help='path where to save, empty for no saving')
    parser.add_argument('--log_dir', default='/output_dir_IN',
                        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('--resume', default='mae_pretrain_vit_base_full.pth',
                        help='resume from checkpoint')
    parser.add_argument('--teacher_path', default='mae_pretrain_vit_large_full.pth',
                        help='resume from checkpoint')

    parser.add_argument('--same_prob', default=1.0, type=float, metavar='N',
                        help='start epoch')
    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.add_argument('--no_pin_mem', action='store_false', dest='pin_mem')
    parser.set_defaults(pin_mem=True)

    # distributed training parameters
    parser.add_argument('--world_size', default=1, type=int,
                        help='number of distributed processes')
    parser.add_argument('--local_rank', default=-1, type=int)
    parser.add_argument('--dist_on_itp', action='store_true')
    parser.add_argument('--dist_url', default='env://',
                        help='url used to set up distributed training')
    
    return parser


def load_pretrained_models(args):
    #misc.init_distributed_mode(args)


    model = models_mae.__dict__[args.model](norm_pix_loss=args.norm_pix_loss)

    #teachermodel
    teach_model = models_mae.__dict__[args.teach_model](norm_pix_loss=args.norm_pix_loss)

    model.cuda()
    teach_model.eval()
    for _, p in teach_model.named_parameters():
        p.requires_grad = False
    teach_model = teach_model.cuda()
    model_without_ddp = model
    teach_without_ddp = teach_model
    print("Model = %s" % str(model_without_ddp))
    
    #teacher
    misc.load_model(args=args, model_without_ddp=teach_without_ddp)
    #student
    checkpoint = torch.load(args.resume, map_location='cpu')

    del checkpoint['optimizer']

    misc.load_model(args=args, model_without_ddp=model_without_ddp,checkpoint=checkpoint)
    return model, teach_model



def main(args):
    
    misc.init_distributed_mode(args)
    
    print('job dir: {}'.format(os.path.dirname(os.path.realpath(__file__))))
    print("{}".format(args).replace(', ', ',\n'))

    device = torch.device(args.device)

    # fix the seed for reproducibility
    seed = args.seed + misc.get_rank()
    torch.manual_seed(seed)
    np.random.seed(seed)

    cudnn.benchmark = True

    # simple augmentation
    transform_train = transforms.Compose([
            transforms.RandomResizedCrop(args.input_size, scale=(0.2, 1.0), interpolation=3),  # 3 is bicubic
            transforms.RandomHorizontalFlip(),
            transforms.ToTensor(),
            transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])])
    #dataset_train = datasets.ImageFolder(os.path.join(args.data_path, 'train'), transform=transform_train)

    dataset_train = classmixINPair(os.path.join(args.data_path, 'train'), same_prob=1.0, transform=transform_train)
    print(dataset_train)

    if True:  # args.distributed:
        num_tasks = misc.get_world_size()
        global_rank = misc.get_rank()
        sampler_train = torch.utils.data.DistributedSampler(
            dataset_train, num_replicas=num_tasks, rank=global_rank, shuffle=True
        )
        print("Sampler_train = %s" % str(sampler_train))
    else:
        sampler_train = torch.utils.data.RandomSampler(dataset_train)

    if global_rank == 0 and args.log_dir is not None:
        os.makedirs(args.log_dir, exist_ok=True)
        log_writer = SummaryWriter(log_dir=args.log_dir)

    else:
        log_writer = None

    data_loader_train = torch.utils.data.DataLoader(
        dataset_train, sampler=sampler_train,
        batch_size=args.batch_size,
        num_workers=args.num_workers,
        pin_memory=args.pin_mem,
        drop_last=True    )
    
    # define the model
    model = models_mae.__dict__[args.model](norm_pix_loss=args.norm_pix_loss)

    #teachermodel
    teach_model = models_mae.__dict__[args.teach_model](norm_pix_loss=args.norm_pix_loss)

    model.to(device)
    teach_model.eval()
    for _, p in teach_model.named_parameters():
        p.requires_grad = False
    teach_model = teach_model.to(device)
    model_without_ddp = model
    teach_without_ddp = teach_model
    print("Model = %s" % str(model_without_ddp))

    eff_batch_size = args.batch_size * args.accum_iter * misc.get_world_size()
    
    if args.lr is None:  # only base_lr is specified
        args.lr = args.blr * eff_batch_size / 256

    print("base lr: %.2e" % (args.lr * 256 / eff_batch_size))
    print("actual lr: %.2e" % args.lr)

    print("accumulate grad iterations: %d" % args.accum_iter)
    print("effective batch size: %d" % eff_batch_size)

    if args.distributed:
        model = torch.nn.parallel.DistributedDataParallel(model, device_ids=[args.gpu], find_unused_parameters=True)
        teach_model = torch.nn.parallel.DistributedDataParallel(teach_model, device_ids=[args.gpu], find_unused_parameters=True)
        teach_without_ddp = teach_model.module
        model_without_ddp = model.module
    
    # following timm: set wd as 0 for bias and norm layers
    param_groups = optim_factory.add_weight_decay(model_without_ddp, args.weight_decay)
    optimizer = torch.optim.AdamW(param_groups, lr=args.lr, betas=(0.9, 0.95))
    print(optimizer)
    loss_scaler = NativeScaler()
    
    #teacher
    misc.load_model(args=args, model_without_ddp=teach_without_ddp)
    #student
    checkpoint = torch.load(args.resume, map_location='cpu')
    checkpoint['model']['decoder_embed1.weight'] = checkpoint['model']['decoder_embed.weight']
    checkpoint['model']['decoder_embed1.bias'] = checkpoint['model']['decoder_embed.bias']
    checkpoint['model']['decoder_embed2.weight'] = checkpoint['model']['decoder_embed.weight']
    checkpoint['model']['decoder_embed2.bias'] = checkpoint['model']['decoder_embed.bias']
    del checkpoint['model']['decoder_embed.weight']
    del checkpoint['model']['decoder_embed.bias']
    misc.load_model(args=args, model_without_ddp=model_without_ddp, optimizer=optimizer, loss_scaler=loss_scaler,checkpoint=checkpoint)

    print(f"Start training for {args.epochs} epochs")
    start_time = time.time()
    for epoch in range(args.start_epoch, args.epochs):
        if args.distributed:
            data_loader_train.sampler.set_epoch(epoch)
        train_stats = train_one_epoch(
            model, data_loader_train,
            optimizer, device, epoch, loss_scaler,
            log_writer=log_writer,
            teach_model=teach_model,
            args=args
        )
        if args.output_dir and (epoch % 5 == 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)

        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")

    total_time = time.time() - start_time
    total_time_str = str(datetime.timedelta(seconds=int(total_time)))
    print('Training time {}'.format(total_time_str))


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