main_lincls.py

#!/usr/bin/env python
# Adapted from MoCo, He, Kaiming, et al. "Momentum contrast for unsupervised visual representation learning."

import os
import time
import json
import torch
import random
import shutil
import warnings
import argparse
import torch.optim
import torch.nn as nn
import torch.utils.data
import torch.nn.parallel
import torch.distributed as dist
from torchvision import datasets
import clipdistiller.models as models
import torch.utils.data.distributed
from tools.dataset import TSVDataset
import torch.backends.cudnn as cudnn
from tools.logger import setup_logger
import torchvision.transforms as transforms
from torch.utils.tensorboard import SummaryWriter

model_names = sorted(name for name in models.__dict__
    if name.islower() and not name.startswith("__")
    and callable(models.__dict__[name]))

parser = argparse.ArgumentParser(description='PyTorch ImageNet Training')
parser.add_argument('--data', metavar='DIR',
                    help='path to dataset')
parser.add_argument('--output', metavar='DIR',
                    help='path to output folder')
parser.add_argument('-a', '--arch', metavar='ARCH', default='resnet50',
                    choices=model_names,
                    help='model architecture: ' +
                        ' | '.join(model_names) +
                        ' (default: resnet50)')
parser.add_argument('-j', '--workers', default=4, type=int, metavar='N',
                    help='number of data loading workers (default: 32)')
parser.add_argument('--epochs', default=100, type=int, metavar='N',
                    help='number of total epochs to run')
parser.add_argument('--start-epoch', default=0, type=int, metavar='N',
                    help='manual epoch number (useful on restarts)')
parser.add_argument('-b', '--batch-size', default=256, type=int,
                    metavar='N',
                    help='mini-batch size (default: 256), this is the total '
                         'batch size of all GPUs')
parser.add_argument('--lr', '--learning-rate', default=30., type=float,
                    metavar='LR', help='initial learning rate', dest='lr')
parser.add_argument('--schedule', default=[60, 80], nargs='*', type=int,
                    help='learning rate schedule (when to drop lr by a ratio)')
parser.add_argument('--momentum', default=0.9, type=float, metavar='M',
                    help='momentum')
parser.add_argument('--wd', '--weight-decay', default=0., type=float,
                    metavar='W', help='weight decay (default: 0.)',
                    dest='weight_decay')
parser.add_argument('-p', '--print-freq', default=100, type=int,
                    metavar='N', help='print frequency (default: 10)')
parser.add_argument('--resume', default='', type=str, metavar='PATH',
                    help='path to latest checkpoint (default: none)')
parser.add_argument('-e', '--evaluate', dest='evaluate', action='store_true',
                    help='evaluate model on validation set')
parser.add_argument('--seed', default=None, type=int,
                    help='seed for initializing training. ')
parser.add_argument('--gpu', default=None, type=int,
                    help='GPU id to use.')

parser.add_argument('--pretrained', default='', type=str,
                    help='path to pretrained checkpoint')

parser.add_argument("--local_rank", type=int, help='local rank for DistributedDataParallel')

best_acc1 = 0


def main():
    args = parser.parse_args()

    if args.seed is not None:
        random.seed(args.seed)
        torch.manual_seed(args.seed)
        cudnn.deterministic = True
        warnings.warn('You have chosen to seed training. '
                      'This will turn on the CUDNN deterministic setting, '
                      'which can slow down your training considerably! '
                      'You may see unexpected behavior when restarting '
                      'from checkpoints.')

    torch.cuda.set_device(args.local_rank)
    torch.distributed.init_process_group(backend='nccl', init_method='env:https://')
    cudnn.benchmark = True

    os.makedirs(args.output, exist_ok=True)
    global logger
    logger = setup_logger(output=args.output, distributed_rank=dist.get_rank(), color=False,
                          name="SEED Linear Evaluation")

    if dist.get_rank() == 0:
        path = os.path.join(args.output, "config.json")
        with open(path, 'w') as f:
            json.dump(vars(args), f, indent=2)
        logger.info("Full config saved to {}".format(path))

    logger.info('world size: {}'.format(dist.get_world_size()))
    logger.info('dist.get_rank(): {}'.format(dist.get_rank()))
    logger.info('local_rank: {}'.format(args.local_rank))

    main_worker(args, logger)


def main_worker(args, logger):
    global best_acc1

    # create model
    logger.info("=> creating model '{}'".format(args.arch))
    model = models.__dict__[args.arch]()

    # freeze all layers but the last fc
    for name, param in model.named_parameters():
        if name not in ['fc.weight', 'fc.bias']:
            param.requires_grad = False
    # init the fc layer
    model.fc.weight.data.normal_(mean=0.0, std=0.01)
    model.fc.bias.data.zero_()

    # load from pre-trained, before DistributedDataParallel constructor
    if args.pretrained:
        if os.path.isfile(args.pretrained):
            logger.info("=> loading checkpoint '{}'".format(args.pretrained))
            checkpoint = torch.load(args.pretrained, map_location="cpu")

            # rename pre-trained keys
            state_dict = checkpoint['state_dict']
            for k in list(state_dict.keys()):
                # retain only encoder student up to before the embedding layer
                if k.startswith('module.student') and not k.startswith('module.student.fc'):
                    # remove prefix
                    state_dict[k[len("module.student."):]] = state_dict[k]

                # delete renamed or unused k
                del state_dict[k]

            msg = model.load_state_dict(state_dict, strict=False)
            assert set(msg.missing_keys) == {"fc.weight", "fc.bias"}

            logger.info("=> loaded pre-trained model '{}'".format(args.pretrained))
        else:
            logger.info("=> no checkpoint found at '{}'".format(args.pretrained))

    model = torch.nn.parallel.DistributedDataParallel(model.cuda(), device_ids=[args.local_rank], broadcast_buffers=False)

    # define loss function (criterion) and optimizer
    criterion = nn.CrossEntropyLoss().cuda()

    # optimize only the linear classifier
    parameters = list(filter(lambda p: p.requires_grad, model.parameters()))
    assert len(parameters) == 2  # fc.weight, fc.bias
    args.lr_mult = args.batch_size / 256
    args.warmup_epochs = 5
    optimizer = torch.optim.SGD(parameters,
                                lr=args.lr_mult * args.lr,
                                momentum=args.momentum,
                                weight_decay=args.weight_decay)

    # optionally resume from a checkpoint
    if args.resume:
        if os.path.isfile(args.resume):
            logger.info("=> loading checkpoint '{}'".format(args.resume))
            checkpoint = torch.load(args.resume)
            args.start_epoch = checkpoint['epoch']
            best_acc1 = checkpoint['best_acc1']
            model.load_state_dict(checkpoint['state_dict'])
            optimizer.load_state_dict(checkpoint['optimizer'])
            logger.info("=> loaded checkpoint '{}' (epoch {})".format(args.resume, checkpoint['epoch']))
        else:
            logger.info("=> no checkpoint found at '{}'".format(args.resume))

    # tensorboard
    if dist.get_rank() == 0:
        summary_writer = SummaryWriter(log_dir=args.output)
    else:
        summary_writer = None

    cudnn.benchmark = True

    # Data loading code
    normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406],
                                     std=[0.229, 0.224, 0.225])

    logger.info("=> loading dataset")
    train_dataset = datasets.ImageFolder(
        os.path.join(args.data, 'train'),
        transform=transforms.Compose([
                    transforms.RandomResizedCrop(224),
                    transforms.RandomHorizontalFlip(),
                    transforms.ToTensor(),
                    normalize,
                ]))

    train_sampler = torch.utils.data.distributed.DistributedSampler(train_dataset)

    assert args.batch_size // dist.get_world_size() == args.batch_size / dist.get_world_size(),\
        'Batch size is not divisible by num of gpus.'
    train_loader = torch.utils.data.DataLoader(
        train_dataset,
        batch_size=args.batch_size // dist.get_world_size(),
        shuffle=(train_sampler is None),
        num_workers=args.workers,
        pin_memory=True,
        sampler=train_sampler)

    val_loader = torch.utils.data.DataLoader(
        datasets.ImageFolder(os.path.join(args.data, 'val'),
                             transforms.Compose([
                                transforms.Resize(256),
                                transforms.CenterCrop(224),
                                transforms.ToTensor(),
                                normalize,
                             ])),
        batch_size=args.batch_size // dist.get_world_size(),
        shuffle=False,
        num_workers=args.workers,
        pin_memory=True)

    logger.info("=> loaded dataset")

    if args.evaluate:
        validate(val_loader, model, criterion, args)
        return

    for epoch in range(args.start_epoch, args.epochs):
        train_sampler.set_epoch(epoch)
        adjust_learning_rate(optimizer, epoch, args)

        # train for one epoch
        logger.info("=> Epoch {} start.".format(epoch))
        train(train_loader, model, criterion, optimizer, epoch, args)

        # evaluate on validation set
        acc1 = validate(val_loader, model, criterion, args)

        # remember best acc@1 and save checkpoint
        is_best = acc1 > best_acc1
        best_acc1 = max(acc1, best_acc1)

        if summary_writer is not None:
            # tensorboard logger
            summary_writer.add_scalar('lincls_acc1', acc1, epoch)
            summary_writer.add_scalar('learning_rate', optimizer.param_groups[0]['lr'], epoch)

        if dist.get_rank() == 0:
            # save_checkpoint({
            #     'epoch': epoch + 1,
            #     'arch': args.arch,
            #     'state_dict': model.state_dict(),
            #     'best_acc1': best_acc1,
            #     'optimizer': optimizer.state_dict(),
            # },
            #     is_best,
            #     filename=os.path.join(args.output, 'lincls_checkpoint_{:04d}.pth.tar'.format(epoch)))
            if epoch == args.start_epoch:
            # if epoch == 0:
                sanity_check(model.state_dict(), args.pretrained)


def train(train_loader, model, criterion, optimizer, epoch, args):
    batch_time = AverageMeter('T', ':5.3f')
    data_time = AverageMeter('DT', ':5.3f')
    losses = AverageMeter('Loss', ':5.3f')
    top1 = AverageMeter('Acc@1', ':5.2f')
    top5 = AverageMeter('Acc@5', ':5.2f')
    lr = ValueMeter('LR', ':5.3f')
    progress = ProgressMeter(
        len(train_loader),
        [batch_time, data_time, lr, losses, top1, top5],
        prefix="Epoch: [{}]".format(epoch))

    def get_learning_rate(optimizer):
        for param_group in optimizer.param_groups:
            return param_group['lr']

    lr.update(get_learning_rate(optimizer))

    """
    Switch to eval mode:
    Under the protocol of linear classification on frozen features/models,
    it is not legitimate to change any part of the pre-trained model.
    BatchNorm in train mode may revise running mean/std (even if it receives
    no gradient), which are part of the model parameters too.
    """
    model.eval()

    end = time.time()
    for i, (images, target) in enumerate(train_loader):
        # measure data loading time
        data_time.update(time.time() - end)

        target = target.cuda()

        # compute output
        output = model(images)
        loss = criterion(output, target)

        # measure accuracy and record loss
        acc1, acc5 = accuracy(output, target, topk=(1, 5))
        losses.update(loss.item(), images.size(0))
        top1.update(acc1[0], images.size(0))
        top5.update(acc5[0], images.size(0))

        # compute gradient and do SGD step
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()

        # measure elapsed time
        batch_time.update(time.time() - end)
        end = time.time()

        if i % args.print_freq == 0:
            progress.display(i)


def validate(val_loader, model, criterion, args):
    batch_time = AverageMeter('Time', ':5.3f')
    losses = AverageMeter('Loss', ':5.3f')
    top1 = AverageMeter('Acc@1', ':5.2f')
    top5 = AverageMeter('Acc@5', ':5.2f')
    progress = ProgressMeter(
        len(val_loader),
        [batch_time, losses, top1, top5],
        prefix='Test: ')

    # switch to evaluate mode
    model.eval()

    with torch.no_grad():
        end = time.time()
        for i, (images, target) in enumerate(val_loader):
            if args.gpu is not None:
                images = images.cuda(args.gpu, non_blocking=True)
            target = target.cuda(args.gpu, non_blocking=True)

            # compute output
            output = model(images)
            loss = criterion(output, target)

            # measure accuracy and record loss
            acc1, acc5 = accuracy(output, target, topk=(1, 5))
            losses.update(loss.item(), images.size(0))
            top1.update(acc1[0], images.size(0))
            top5.update(acc5[0], images.size(0))

            # measure elapsed time
            batch_time.update(time.time() - end)
            end = time.time()

            if i % args.print_freq == 0:
                progress.display(i)

        # TODO: this should also be done with the ProgressMeter
        logger.info(' * Acc@1 {top1.avg:.3f} Acc@5 {top5.avg:.3f}'.format(top1=top1, top5=top5))

    return top1.avg


def save_checkpoint(state, is_best, filename='checkpoint.pth.tar'):
    torch.save(state, filename)
    if is_best:
        shutil.copyfile(filename, 'model_best.pth.tar')


def sanity_check(state_dict, pretrained_weights):
    """
    Linear classifier should not change any weights other than the linear layer.
    This sanity check asserts nothing wrong happens (e.g., BN stats updated).
    """
    logger.info("=> loading '{}' for sanity check".format(pretrained_weights))
    checkpoint = torch.load(pretrained_weights, map_location="cpu")
    state_dict_pre = checkpoint['state_dict']

    for k in list(state_dict.keys()):
        # only ignore fc layer
        if 'fc.weight' in k or 'fc.bias' in k:
            continue

        # name in pretrained model
        k_pre = 'module.student.' + k[len('module.'):] if k.startswith('module.') else 'module.student.' + k

        assert ((state_dict[k].cpu() == state_dict_pre[k_pre]).all()), \
            '{} is changed in linear classifier training.'.format(k)

    logger.info("=> sanity check passed.")


class AverageMeter(object):
    """Computes and stores the average and current value"""
    def __init__(self, name, fmt=':f'):
        self.name = name
        self.fmt = fmt
        self.reset()

    def reset(self):
        self.val = 0
        self.avg = 0
        self.sum = 0
        self.count = 0

    def update(self, val, n=1):
        self.val = val
        self.sum += val * n
        self.count += n
        self.avg = self.sum / self.count

    def __str__(self):
        fmtstr = '{name} {val' + self.fmt + '} ({avg' + self.fmt + '})'
        return fmtstr.format(**self.__dict__)


class ValueMeter(object):
    """stores the current value"""
    def __init__(self, name, fmt=':f'):
        self.name = name
        self.fmt = fmt
        self.reset()

    def reset(self):
        self.val = 0

    def update(self, val):
        self.val = val

    def __str__(self):
        fmtstr = '{name} {val' + self.fmt + '}'
        return fmtstr.format(**self.__dict__)


class ProgressMeter(object):
    def __init__(self, num_batches, meters, prefix=""):
        self.batch_fmtstr = self._get_batch_fmtstr(num_batches)
        self.meters = meters
        self.prefix = prefix

    def display(self, batch):
        entries = [self.prefix + self.batch_fmtstr.format(batch)]
        entries += [str(meter) for meter in self.meters]
        logger.info('  '.join(entries))

    def _get_batch_fmtstr(self, num_batches):
        num_digits = len(str(num_batches // 1))
        fmt = '{:' + str(num_digits) + 'd}'
        return '[' + fmt + '/' + fmt.format(num_batches) + ']'


def adjust_learning_rate(optimizer, epoch, args):
    """Decay the learning rate based on schedule"""
    lr = args.lr * args.lr_mult
    if epoch < args.warmup_epochs:
        # warm up
        lr = args.lr + (args.lr * args.lr_mult - args.lr) / args.warmup_epochs * epoch

    for milestone in args.schedule:
        lr *= 0.1 if epoch >= milestone else 1.
    for param_group in optimizer.param_groups:
        param_group['lr'] = lr


def accuracy(output, target, topk=(1,)):
    """Computes the accuracy over the k top predictions for the specified values of k"""
    with torch.no_grad():
        maxk = max(topk)
        batch_size = target.size(0)

        _, pred = output.topk(maxk, 1, True, True)
        pred = pred.t()
        correct = pred.eq(target.view(1, -1).expand_as(pred))

        res = []
        for k in topk:
            correct_k = correct[:k].contiguous().view(-1).float().sum(0, keepdim=True)
            res.append(correct_k.mul_(100.0 / batch_size))
        return res


if __name__ == '__main__':
    main()