train_kmeans.py

import math
import os.path
import argparse
import os
import random
import shutil
import time
import warnings
import sys
from collections import OrderedDict

import torch
import torch.nn as nn
import torch.nn.parallel
import torch.backends.cudnn as cudnn
import torch.optim
import torch.utils.data
from torch.utils.data import DataLoader
import torchvision.transforms as transforms
import torchvision.datasets as datasets
import torchvision.models as models
import torch.nn.functional as F

from PIL import Image

from tools import *


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='Unsupervised distillation')
parser.add_argument('data', metavar='DIR',
                    help='path to dataset')
parser.add_argument('-j', '--workers', default=4, type=int, metavar='N',
                    help='number of data loading workers (default: 4)')
parser.add_argument('-a', '--arch', default='resnet18',
                    help='model architecture: ' +
                         ' | '.join(model_names) +
                         ' (default: resnet18)')
parser.add_argument('--epochs', default=90, 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 on the current node when '
                         'using Data Parallel or Distributed Data Parallel')
parser.add_argument('--lr', '--learning-rate', default=0.1, type=float,
                    help='initial learning rate', dest='lr')
parser.add_argument('--cos', action='store_true',
                    help='whether to cosine learning rate or not')
parser.add_argument('--schedule', type=int, nargs='*',
                    default=[30,60,80,90],
                    help='lr drop schedule')
parser.add_argument('--momentum', default=0.9, type=float, metavar='M',
                    help='momentum')
parser.add_argument('--wd', '--weight-decay', default=1e-4, type=float,
                    metavar='W', help='weight decay (default: 1e-4)',
                    dest='weight_decay')
parser.add_argument('-p', '--print-freq', default=90, 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('--seed', default=None, type=int,
                    help='seed for initializing training. ')
parser.add_argument('--save', default='./output/distill_1', type=str,
                    help='experiment output directory')
parser.add_argument('-e', '--evaluate', dest='evaluate', action='store_true',
                    help='evaluate model on validation set')
parser.add_argument('--clusters', type=str, required=True,
                    help='location containing cluster assignment files')

best_acc1 = 0


class ImageFileDataset(datasets.VisionDataset):
    def __init__(self, root, f_path, transform=None):
        super(ImageFileDataset, self).__init__(root, transform=transform)
        with open(f_path, 'r') as f:
            lines = [line.strip().split(' ') for line in f.readlines()]
            lines = [(os.path.join(root, pth), int(cid)) for pth, cid in lines]
        self.samples = lines
        self.classes = sorted(set(s[1] for s in self.samples))

    def __getitem__(self, index):
        path, target = self.samples[index]
        sample = default_loader(path)
        sample = self.transform(sample)

        return sample, target

    def __len__(self):
        return len(self.samples)


def pil_loader(path):
    with open(path, 'rb') as f:
        img = Image.open(f)
        return img.convert('RGB')


def accimage_loader(path):
    import accimage
    try:
        return accimage.Image(path)
    except IOError:
        # Potentially a decoding problem, fall back to PIL.Image
        return pil_loader(path)


def default_loader(path):
    from torchvision import get_image_backend
    if get_image_backend() == 'accimage':
        return accimage_loader(path)
    else:
        return pil_loader(path)


def main():
    global logger

    args = parser.parse_args()
    makedirs(args.save)
    logger = get_logger(logpath=os.path.join(args.save, 'logs'), filepath=os.path.abspath(__file__))
    logger.info(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.')

    main_worker(args)


def main_worker(args):
    global best_acc1

    # Data loading code
    traindir = os.path.join(args.data, 'train')
    valdir = os.path.join(args.data, 'val')
    normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406],
                                     std=[0.229, 0.224, 0.225])
    train_clusters = os.path.join(args.clusters, 'train_clusters.txt')
    val_clusters = os.path.join(args.clusters, 'val_clusters.txt')

    train_dataset = ImageFileDataset(
        traindir,
	train_clusters,
        transforms.Compose([
            transforms.RandomResizedCrop(224),
            transforms.RandomHorizontalFlip(),
            transforms.ToTensor(),
            normalize,
        ]))

    train_loader = DataLoader(
        train_dataset, batch_size=args.batch_size, shuffle=True,
        num_workers=args.workers, pin_memory=True)

    val_loader = torch.utils.data.DataLoader(
        ImageFileDataset(
            valdir,
            val_clusters,
            transforms.Compose([
                transforms.Resize(256),
                transforms.CenterCrop(224),
                transforms.ToTensor(),
                normalize,
            ])
        ),
        batch_size=args.batch_size, shuffle=False,
        num_workers=args.workers, pin_memory=True)

    model = models.__dict__[args.arch](num_classes=len(train_dataset.classes))
    model = nn.DataParallel(model).cuda()

    bn_params = OrderedDict()
    for n, m in model.named_modules():
        if isinstance(m, nn.BatchNorm2d):
            for p in m.parameters():
                bn_params[p] = True
    params = OrderedDict()
    for p in model.parameters():
        if p not in bn_params:
            params[p] = True
    param_groups = [
            {'params': list(params.keys())},
            {'params': list(bn_params.keys()), 'weight_decay': 0},
    ]

    optimizer = torch.optim.SGD(param_groups,
                                args.lr,
                                momentum=args.momentum,
                                weight_decay=args.weight_decay,
                                nesterov=True)

    # 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']
            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))

    cudnn.benchmark = True

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

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

        # train for one epoch
        train(train_loader, model, optimizer, epoch, args)

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

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

        save_checkpoint({
            'epoch': epoch + 1,
            'state_dict': model.state_dict(),
            'best_acc1': best_acc1,
            'optimizer': optimizer.state_dict(),
        }, is_best, args.save)


def train(train_loader, model, optimizer, epoch, args):
    batch_time = AverageMeter('Time', ':6.3f')
    data_time = AverageMeter('Data', ':6.3f')
    losses = AverageMeter('Loss', ':.4e')
    top1 = AverageMeter('Acc@1', ':6.2f')
    top5 = AverageMeter('Acc@5', ':6.2f')
    progress = ProgressMeter(
        len(train_loader),
        [batch_time, data_time, losses, top1, top5],
        prefix="Epoch: [{}]".format(epoch))

    # switch to train mode
    model.train()

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

        images = images.cuda(non_blocking=True)
        target = target.cuda(non_blocking=True)

        # compute output
        output = model(images)
        loss = F.cross_entropy(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:
            logger.info(progress.display(i))


def validate(val_loader, model, args):
    batch_time = AverageMeter('Time', ':6.3f')
    losses = AverageMeter('Loss', ':.4e')
    top1 = AverageMeter('Acc@1', ':6.2f')
    top5 = AverageMeter('Acc@5', ':6.2f')
    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):
            images = images.cuda(non_blocking=True)
            target = target.cuda(non_blocking=True)

            # compute output
            output = model(images)
            loss = F.cross_entropy(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:
                logger.info(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 adjust_learning_rate(optimizer, epoch, args):
    """Decay the learning rate based on schedule"""
    lr = args.lr
    if args.cos:  # cosine lr schedule
        lr *= 0.5 * (1. + math.cos(math.pi * epoch / args.epochs))
    else:  # stepwise lr schedule
        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 print_lr(optimizer):
    lrs = [param_group['lr'] for param_group in optimizer.param_groups]
    lrs = ' '.join('{:f}'.format(l) for l in lrs)
    logger.info('LR: ' + lrs)


if __name__ == '__main__':
    main()