main_byol.py

# some code in this file is adapted from
# https://github.com/pytorch/examples
# Original Copyright 2017. Licensed under the BSD 3-Clause License.
# Modifications Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
# SPDX-License-Identifier: CC-BY-NC-4.0

import argparse
import builtins
import os
import time
import warnings
import numpy as np

import torch
import torch.nn.parallel
import torch.backends.cudnn as cudnn
import torch.distributed as dist
import torch.optim
import torch.multiprocessing as mp
import torch.utils.data
import torch.utils.data.distributed

import data.datasets as custom_datasets
import torchvision.datasets as datasets
import backbone as backbone_models
from models import get_byol_model
from utils import utils, lr_schedule, LARS, get_norm, dist_utils
import data.transforms as data_transforms
from engine import ss_validate

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

parser = argparse.ArgumentParser(description='PyTorch ImageNet Training')
parser.add_argument('data', metavar='DIR',
                    help='path to dataset')
parser.add_argument('--arch', metavar='ARCH', default='BYOL',
                    help='model architecture')
parser.add_argument('--backbone', default='resnet50_encoder',
                    choices=backbone_model_names,
                    help='model architecture: ' +
                        ' | '.join(backbone_model_names) +
                        ' (default: resnet50_encoder)')
parser.add_argument('-j', '--workers', default=32, type=int, metavar='N',
                    help='number of data loading workers (default: 32)')
parser.add_argument('--epochs', default=200, 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('--warmup-epoch', default=0, type=int, metavar='N',
                    help='number of epochs for learning warmup')
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=0.03, type=float,
                    metavar='LR', help='initial learning rate', dest='lr')
parser.add_argument('--schedule', default=[120, 160], nargs='*', type=int,
                    help='learning rate schedule (when to drop lr by 10x)')
parser.add_argument('--cos', action='store_true', help='use cosine lr schedule')
parser.add_argument('--momentum', default=0.9, type=float, metavar='M',
                    help='momentum of SGD solver')
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=100, type=int,
                    metavar='N', help='print frequency (default: 10)')
parser.add_argument('--save-freq', default=50, type=int,
                    metavar='N', help='checkpoint save frequency (default: 1)')
parser.add_argument('--eval-freq', default=5, type=int,
                    metavar='N', help='evaluation epoch frequency (default: 5)')
parser.add_argument('--resume', default='', type=str, metavar='PATH',
                    help='path to latest checkpoint (default: none)')
parser.add_argument('--pretrained', default='', type=str, metavar='PATH',
                    help='path to pretrained model (default: none)')
parser.add_argument('--super-pretrained', default='', type=str, metavar='PATH',
                    help='path to MoCo pretrained model (default: none)')
parser.add_argument('-e', '--evaluate', dest='evaluate', action='store_true',
                    help='evaluate model on validation set')
parser.add_argument('--world-size', default=-1, type=int,
                    help='number of nodes for distributed training')
parser.add_argument('--rank', default=-1, type=int,
                    help='node rank for distributed training')
parser.add_argument('--dist-url', default='tcp://224.66.41.62:23456', type=str,
                    help='url used to set up distributed training')
parser.add_argument('--dist-backend', default='nccl', type=str,
                    help='distributed backend')
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('--multiprocessing-distributed', action='store_true',
                    help='Use multi-processing distributed training to launch '
                         'N processes per node, which has N GPUs. This is the '
                         'fastest way to use PyTorch for either single node or '
                         'multi node data parallel training')

# byol specific configs:
parser.add_argument('--byol-dim', default=256, type=int,
                    help='feature dimension (default: 128)')
parser.add_argument('--byol-m', default=0.996, type=float,
                    help='moco momentum of updating key encoder (default: 0.999)')
parser.add_argument('--norm', default='None', type=str,
                    help='the normalization for network (default: None)')
parser.add_argument('--num-neck-mlp', default=2, type=int,
                    help='number of neck mlp (default: 2)')
parser.add_argument('--hid-dim', default=4096, type=int,
                    help='hidden dimension of mlp (default: 4096)')

# options for KNN search
parser.add_argument('--num-nn', default=20, type=int,
                    help='Number of nearest neighbors (default: 20)')
parser.add_argument('--nn-mem-percent', type=float, default=0.1,
                    help='number of percentage mem datan for KNN evaluation')
parser.add_argument('--nn-query-percent', type=float, default=0.5,
                    help='number of percentage query datan for KNN evaluation')


best_acc1 = 0


def main():
    args = parser.parse_args()
    assert args.warmup_epoch < args.schedule[0]
    print(args)

    if args.seed is not None:
        seed = args.seed + dist_utils.get_rank()
        torch.manual_seed(seed)
        np.random.seed(seed)
        # random.seed(seed)

    if args.gpu is not None:
        warnings.warn('You have chosen a specific GPU. This will completely '
                      'disable data parallelism.')

    if args.dist_url == "env://" and args.world_size == -1:
        args.world_size = int(os.environ["WORLD_SIZE"])

    args.distributed = args.world_size > 1 or args.multiprocessing_distributed

    ngpus_per_node = torch.cuda.device_count()
    if args.multiprocessing_distributed:
        # Since we have ngpus_per_node processes per node, the total world_size
        # needs to be adjusted accordingly
        args.world_size = ngpus_per_node * args.world_size
        # Use torch.multiprocessing.spawn to launch distributed processes: the
        # main_worker process function
        mp.spawn(main_worker, nprocs=ngpus_per_node, args=(ngpus_per_node, args))
    else:
        # Simply call main_worker function
        main_worker(args.gpu, ngpus_per_node, args)


def main_worker(gpu, ngpus_per_node, args):
    global best_acc1
    args.gpu = gpu

    # suppress printing if not master
    if args.multiprocessing_distributed and args.gpu != 0:
        def print_pass(*args):
            pass
        builtins.print = print_pass

    if args.gpu is not None:
        print("Use GPU: {} for training".format(args.gpu))

    if args.distributed:
        if args.dist_url == "env://" and args.rank == -1:
            args.rank = int(os.environ["RANK"])
        if args.multiprocessing_distributed:
            # For multiprocessing distributed training, rank needs to be the
            # global rank among all the processes
            args.rank = args.rank * ngpus_per_node + gpu
        dist.init_process_group(backend=args.dist_backend, init_method=args.dist_url,
                                world_size=args.world_size, rank=args.rank)

    # create model
    print("=> creating model '{}' with backbone '{}'".format(args.arch, args.backbone))
    model_func = get_byol_model(args.arch)
    norm_layer = get_norm(args.norm)
    model = model_func(
        backbone_models.__dict__[args.backbone],
        dim=args.byol_dim,
        m=args.byol_m,
        hid_dim=args.hid_dim,
        norm_layer=norm_layer,
        num_neck_mlp=args.num_neck_mlp,
    )
    print(model)

    if args.pretrained:
        if os.path.isfile(args.pretrained):
            print("=> loading pretrained model from '{}'".format(args.pretrained))
            state_dict = torch.load(args.pretrained, map_location="cpu")['state_dict']
            # rename state_dict keys
            for k in list(state_dict.keys()):
                new_key = k.replace("module.", "")
                state_dict[new_key] = state_dict[k]
                del state_dict[k]
            msg = model.load_state_dict(state_dict, strict=False)
            print("=> loaded pretrained model from '{}'".format(args.pretrained))
            if len(msg.missing_keys) > 0:
                print("missing keys: {}".format(msg.missing_keys))
            if len(msg.unexpected_keys) > 0:
                print("unexpected keys: {}".format(msg.unexpected_keys))
        else:
            print("=> no pretrained model found at '{}'".format(args.pretrained))

    if args.distributed:
        # For multiprocessing distributed, DistributedDataParallel constructor
        # should always set the single device scope, otherwise,
        # DistributedDataParallel will use all available devices.
        if args.gpu is not None:
            torch.cuda.set_device(args.gpu)
            model.cuda(args.gpu)
            # When using a single GPU per process and per
            # DistributedDataParallel, we need to divide the batch size
            # ourselves based on the total number of GPUs we have
            args.batch_size = int(args.batch_size / ngpus_per_node)
            args.workers = int((args.workers + ngpus_per_node - 1) / ngpus_per_node)
            model = torch.nn.parallel.DistributedDataParallel(model, device_ids=[args.gpu])
        else:
            model.cuda()
            # DistributedDataParallel will divide and allocate batch_size to all
            # available GPUs if device_ids are not set
            model = torch.nn.parallel.DistributedDataParallel(model)
    elif args.gpu is not None:
        torch.cuda.set_device(args.gpu)
        model = model.cuda(args.gpu)
    else:
        model = torch.nn.DataParallel(model).cuda()

    # define optimizer
    params = collect_params(model, exclude_bias_and_bn=True)
    optimizer = LARS(params, lr=args.lr, momentum=args.momentum, weight_decay=args.weight_decay)

    # optionally resume from a checkpoint
    if args.resume:
        if os.path.isfile(args.resume):
            print("=> loading checkpoint '{}'".format(args.resume))
            if args.gpu is None:
                checkpoint = torch.load(args.resume)
            else:
                # Map model to be loaded to specified single gpu.
                loc = 'cuda:{}'.format(args.gpu)
                checkpoint = torch.load(args.resume, map_location=loc)
            args.start_epoch = checkpoint['epoch']
            if 'best_acc1' in checkpoint:
                best_acc1 = checkpoint['best_acc1']
                #if args.gpu is not None:
                #    # best_acc1 may be from a checkpoint from a different GPU
                #    best_acc1 = best_acc1.to(args.gpu)
            model.load_state_dict(checkpoint['state_dict'])
            optimizer.load_state_dict(checkpoint['optimizer'])
            print("=> loaded checkpoint '{}' (epoch {})"
                  .format(args.resume, checkpoint['epoch']))
        else:
            print("=> no checkpoint found at '{}'".format(args.resume))

    cudnn.benchmark = True

    # Data loading code
    traindir = os.path.join(args.data, 'train')
    valdir = os.path.join(args.data, 'val')

    transform1, transform2 = data_transforms.get_byol_tranforms()
    train_dataset = datasets.ImageFolder(
        traindir,
        data_transforms.TwoCropsTransform(transform1, transform2))
    print("train_dataset:\n{}".format(train_dataset))

    if args.distributed:
        train_sampler = torch.utils.data.distributed.DistributedSampler(train_dataset)
    else:
        train_sampler = None

    train_loader = torch.utils.data.DataLoader(
        train_dataset, batch_size=args.batch_size, shuffle=(train_sampler is None),
        num_workers=args.workers, pin_memory=True, sampler=train_sampler, drop_last=True)

    val_loader_base = torch.utils.data.DataLoader(
        custom_datasets.ImageFolderWithPercent(
            traindir,
            data_transforms.get_transforms("DefaultVal"),
            percent=args.nn_mem_percent
        ),
        batch_size=args.batch_size, shuffle=False,
        num_workers=args.workers, pin_memory=True)

    val_loader_query = torch.utils.data.DataLoader(
        custom_datasets.ImageFolderWithPercent(
            valdir,
            data_transforms.get_transforms("DefaultVal"),
            percent=args.nn_query_percent
        ),
        batch_size=args.batch_size, shuffle=False,
        num_workers=args.workers, pin_memory=True)

    if args.evaluate:
        ss_validate(val_loader_base, val_loader_query, model, args)
        return

    best_epoch = args.start_epoch
    for epoch in range(args.start_epoch, args.epochs):
        if args.distributed:
            train_sampler.set_epoch(epoch)
        if epoch >= args.warmup_epoch:
            lr_schedule.adjust_learning_rate(optimizer, epoch, args)

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

        is_best = False
        if (epoch + 1) % args.eval_freq == 0:
            acc1 = ss_validate(val_loader_base, val_loader_query, model, args)
            # remember best acc@1 and save checkpoint
            is_best = acc1 > best_acc1
            best_acc1 = max(acc1, best_acc1)
            if is_best:
                best_epoch = epoch

        if not args.multiprocessing_distributed or (args.multiprocessing_distributed
                and args.rank % ngpus_per_node == 0):
            utils.save_checkpoint({
                'epoch': epoch + 1,
                'arch': args.arch,
                'state_dict': model.state_dict(),
                'best_acc1': best_acc1,
                'optimizer': optimizer.state_dict(),
            }, is_best=is_best, epoch=epoch, args=args)

    print('Best Acc@1 {0} @ epoch {1}'.format(best_acc1, best_epoch + 1))


def train(train_loader, model, optimizer, epoch, args):
    batch_time = utils.AverageMeter('Time', ':6.3f')
    data_time = utils.AverageMeter('Data', ':6.3f')
    losses = utils.AverageMeter('Loss', ':.4e')
    curr_lr = utils.InstantMeter('LR', '')
    curr_mom = utils.InstantMeter('MOM', '')
    progress = utils.ProgressMeter(
        len(train_loader),
        [curr_lr, curr_mom, batch_time, data_time, losses],
        prefix="Epoch: [{}/{}]\t".format(epoch, args.epochs))

    # iter info
    batch_iter = len(train_loader)
    max_iter = float(batch_iter * args.epochs)

    # switch to train mode
    model.train()
    if "EMAN" in args.arch:
        print("setting the key model to eval mode when using EMAN")
        if hasattr(model, 'module'):
            model.module.target_net.eval()
        else:
            model.target_net.eval()

    end = time.time()
    for i, (images, _) in enumerate(train_loader):
        # update model momentum
        curr_iter = float(epoch * batch_iter + i)

        # measure data loading time
        data_time.update(time.time() - end)

        if args.gpu is not None:
            images[0] = images[0].cuda(args.gpu, non_blocking=True)
            images[1] = images[1].cuda(args.gpu, non_blocking=True)

        # warmup learning rate
        if epoch < args.warmup_epoch:
            warmup_step = args.warmup_epoch * batch_iter
            curr_step = epoch * batch_iter + i + 1
            lr_schedule.warmup_learning_rate(optimizer, curr_step, warmup_step, args)
        curr_lr.update(optimizer.param_groups[0]['lr'])

        # compute loss
        loss = model(im_v1=images[0], im_v2=images[1])

        # measure accuracy and record loss
        losses.update(loss.item(), images[0].size(0))

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

        # update the target model
        if hasattr(model, 'module'):
            model.module.momentum_update(curr_iter, max_iter)
            curr_mom.update(model.module.curr_m)
        else:
            model.momentum_update(curr_iter, max_iter)
            curr_mom.update(model.curr_m)

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

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


def collect_params(model, exclude_bias_and_bn=True):
    """
    exclude_bias_and bn: exclude bias and bn from both weight decay and LARS adaptation
        in the PyTorch implementation of ResNet, `downsample.1` are bn layers
    """
    weight_param_list, bn_and_bias_param_list = [], []
    weight_param_names, bn_and_bias_param_names = [], []
    for name, param in model.named_parameters():
        if exclude_bias_and_bn and ('bn' in name or 'downsample.1' in name or 'bias' in name):
            bn_and_bias_param_list.append(param)
            bn_and_bias_param_names.append(name)
        else:
            weight_param_list.append(param)
            weight_param_names.append(name)
    print("weight params:\n{}".format('\n'.join(weight_param_names)))
    print("bn and bias params:\n{}".format('\n'.join(bn_and_bias_param_names)))
    param_list = [{'params': bn_and_bias_param_list, 'weight_decay': 0., 'lars_exclude': True},
                  {'params': weight_param_list}]
    return param_list


if __name__ == '__main__':
    main()