main_simsiam.py

#!/usr/bin/env python
# Copyright (c) Facebook, Inc. and its affiliates.
# All rights reserved.

# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.

import argparse
import builtins
import contextlib
import math
import os
import random
import shutil
import time
import warnings

import torch
import torch.backends.cudnn as cudnn
import torch.distributed as dist
import torch.multiprocessing as mp
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
import torch.utils.data.distributed
import torchvision.datasets as datasets
import torchvision.models as models
import torchvision.transforms as transforms

import simsiam.builder
import simsiam.loader

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(
    "-a",
    "--arch",
    metavar="ARCH",
    default="resnet50",
    choices=model_names,
    help="model architecture: " + " | ".join(model_names) + " (default: resnet50)",
)
parser.add_argument(
    "-j",
    "--workers",
    default=32,
    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=512,
    type=int,
    metavar="N",
    help="mini-batch size (default: 512), 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=0.05,
    type=float,
    metavar="LR",
    help="initial (base) learning rate",
    dest="lr",
)
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=10,
    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(
    "--checkpoint-dir",
    default=".",
    type=str,
    metavar="PATH",
    help="output checkpoint directory (default: current directory)",
)
parser.add_argument(
    "--num-checkpoints",
    default=3,
    type=int,
    help="number of past checkpoints to keep (default: 3)",
)
parser.add_argument(
    "--min-checkpoint-interval",
    default=600,
    type=float,
    help=(
        "minimum amount of time elapsed between saving checkpoints,"
        " in seconds (default: 600s = 10min)"
    ),
)
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",
)

# simsiam specific configs:
parser.add_argument(
    "--dim", default=2048, type=int, help="feature dimension (default: 2048)"
)
parser.add_argument(
    "--pred-dim",
    default=512,
    type=int,
    help="hidden dimension of the predictor (default: 512)",
)
parser.add_argument(
    "--fix-pred-lr", action="store_true", help="Fix learning rate for the predictor"
)

# byol arguments
parser.add_argument(
    "--ema",
    nargs="?",
    const=0.99,
    type=float,
    metavar="ALPHA",
    help=(
        "Momentum value for exponential moving average updates to teacher"
        " model (BYOL model)."
        " If --ema is supplied without specifying ALPHA, the default ALPHA"
        " value of %(const)s is used."
        " If this argument is not supplied (default), a SimSiam model is"
        " trained instead."
    ),
)
parser.add_argument(
    "--ema-init-target-from-online",
    action="store_true",
    help="Initialize target network weights from online weights",
)


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."
        )

    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):
    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,
        )
        torch.distributed.barrier()
    # create model
    print("=> creating model '{}'".format(args.arch))
    if args.ema is not None:
        if args.ema_init_target_from_online:
            init_str = "online"
        else:
            init_str = "scratch"
        print(
            "   Using BYOL model with EMA alpha={}, target init from {},"
            " dim={}, pred_dim={}".format(args.ema, init_str, args.dim, args.pred_dim)
        )
        model = simsiam.builder.BYOL(
            models.__dict__[args.arch],
            dim=args.dim,
            pred_dim=args.pred_dim,
            init_target_from_online=args.ema_init_target_from_online,
            alpha=args.ema,
        )
    else:
        print(
            "   Using SimSiam model, dim={}, pred_dim={}".format(
                args.dim, args.pred_dim
            )
        )
        model = simsiam.builder.SimSiam(
            models.__dict__[args.arch], dim=args.dim, pred_dim=args.pred_dim
        )

    # infer learning rate before changing batch size
    init_lr = args.lr * args.batch_size / 256

    if not torch.cuda.is_available():
        print("using CPU, this will be slow")
    elif args.distributed:
        # apply SyncBN
        model = torch.nn.SyncBatchNorm.convert_sync_batchnorm(model)
        # 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 / args.world_size)
            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)
        # comment out the following line for debugging
        raise NotImplementedError("Only DistributedDataParallel is supported.")
    else:
        # AllGather implementation (batch shuffle, queue update, etc.) in
        # this code only supports DistributedDataParallel.
        raise NotImplementedError("Only DistributedDataParallel is supported.")
    print(model)  # print model after SyncBatchNorm

    # define loss function (criterion) and optimizer
    criterion = nn.CosineSimilarity(dim=1).cuda(args.gpu)

    if args.fix_pred_lr:
        optim_params = [
            {"params": model.module.encoder.parameters(), "fix_lr": False},
            {"params": model.module.predictor.parameters(), "fix_lr": True},
        ]
    else:
        optim_params = model.parameters()

    optimizer = torch.optim.SGD(
        optim_params, init_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"]
            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")
    normalize = transforms.Normalize(
        mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]
    )

    # MoCo v2's aug: similar to SimCLR https://arxiv.org/abs/2002.05709
    augmentation = [
        transforms.RandomResizedCrop(224, scale=(0.2, 1.0)),
        transforms.RandomApply(
            [transforms.ColorJitter(0.4, 0.4, 0.4, 0.1)], p=0.8  # not strengthened
        ),
        transforms.RandomGrayscale(p=0.2),
        transforms.RandomApply([simsiam.loader.GaussianBlur([0.1, 2.0])], p=0.5),
        transforms.RandomHorizontalFlip(),
        transforms.ToTensor(),
        normalize,
    ]

    train_dataset = datasets.ImageFolder(
        traindir, simsiam.loader.TwoCropsTransform(transforms.Compose(augmentation))
    )

    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,
    )

    checkpoint_paths = []
    t_last_checkpoint = None

    for epoch in range(args.start_epoch + 1, args.epochs + 1):
        if args.distributed:
            train_sampler.set_epoch(epoch)
        adjust_learning_rate(optimizer, init_lr, epoch, args)

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

        if not args.multiprocessing_distributed or (
            args.multiprocessing_distributed and args.rank % ngpus_per_node == 0
        ):
            ckpt_fmt = "checkpoint_{:04d}.pt"
            ckpt_path = os.path.join(args.checkpoint_dir, ckpt_fmt.format(epoch))
            if (
                t_last_checkpoint is None
                or epoch >= args.epochs - 1
                or time.time() - t_last_checkpoint >= args.min_checkpoint_interval
            ):
                save_checkpoint(
                    {
                        "epoch": epoch,
                        "arch": args.arch,
                        "state_dict": model.state_dict(),
                        "optimizer": optimizer.state_dict(),
                    },
                    is_best=False,
                    filename=ckpt_path,
                )
                checkpoint_paths.append(ckpt_path)
                t_last_checkpoint = time.time()
                # Make checkpoint_latest be a symbolic link to the new checkpoint.
                # Since the link will already exist, we make a temporary symbolic
                # link and copy it over to overwrite the destination.
                tmp_link = ckpt_path + ".{}.tmp".format(time.time())
                os.symlink(ckpt_path, tmp_link)
                os.rename(
                    tmp_link,
                    os.path.join(args.checkpoint_dir, "checkpoint_latest.pt"),
                )
            # Remove an old checkpoint, to save space
            if args.num_checkpoints <= 0:
                # Never remove old checkpoints if we are trying to store them
                # all
                pass
            elif len(checkpoint_paths) > args.num_checkpoints:
                # Remove an old checkpoint which was saved in this run to make
                # space
                ckpt_path_old = checkpoint_paths.pop(0)
                with contextlib.suppress(FileNotFoundError):
                    os.remove(ckpt_path_old)
            elif len(checkpoint_paths) > 0:
                # To make sure space is cleared, we remove checkpoints which
                # may have been created by previous runs of this script
                # (before it was pre-empted or otherwise restarted)
                ckpt_path_old = os.path.join(
                    args.checkpoint_dir,
                    ckpt_fmt.format(epoch - args.num_checkpoints),
                )
                if ckpt_path_old not in checkpoint_paths:
                    with contextlib.suppress(FileNotFoundError):
                        os.remove(ckpt_path_old)


def train(train_loader, model, criterion, optimizer, epoch, args):
    batch_time = AverageMeter("Time", ":6.3f")
    data_time = AverageMeter("Data", ":6.3f")
    losses = AverageMeter("Loss", ":.4f")
    progress = ProgressMeter(
        len(train_loader),
        [batch_time, data_time, losses],
        prefix="Epoch: [{}]".format(epoch),
    )

    # switch to train mode
    model.train()

    end = time.time()
    for i, (images, _) in enumerate(train_loader):
        # 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)

        # compute output and loss
        p1, p2, z1, z2 = model(x1=images[0], x2=images[1])
        loss = -(criterion(p1, z2).mean() + criterion(p2, z1).mean()) * 0.5

        losses.update(loss.item(), images[0].size(0))

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

        # update target model if BYOL
        if args.ema:
            model.module.update_target()

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

        if i % args.print_freq == 0 or i <= 2:
            progress.display(i)


def save_checkpoint(state, is_best, filename="checkpoint.pt"):
    print("Saving checkpoint: {}".format(filename))
    torch.save(state, filename)
    if is_best:
        shutil.copyfile(filename, "model_best.pt")


class AverageMeter(object):
    """Compute and store 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 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]
        print("\t".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, init_lr, epoch, args):
    """Decay the learning rate based on schedule."""
    cur_lr = init_lr * 0.5 * (1.0 + math.cos(math.pi * epoch / args.epochs))
    for param_group in optimizer.param_groups:
        if "fix_lr" in param_group and param_group["fix_lr"]:
            param_group["lr"] = init_lr
        else:
            param_group["lr"] = cur_lr


if __name__ == "__main__":
    main()