#!/usr/bin/env python # -*- coding: utf-8 -*- # @Date : 2021-01-27 15:38:58 # @Author : Dengpan Fu (t-defu@microsoft.com) import argparse import builtins import math import os, sys import random import shutil import time import warnings import numpy as np import torch import torch.nn as nn 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 torchvision.transforms as transforms import torchvision.datasets as datasets import torchvision.models as models from libs.dataset import PersonDataset, CMDM from libs.transform import get_lup_transformer, get_reid_test_transformer from libs.moco_builder import MoCo from libs.loader import IterDistributedSampler, InferenceSampler from libs.lars import get_lars_optimizer from libs.utils import gather_tensors, compute_reid_score, print_scores try: from torch.cuda.amp import autocast, GradScaler MIX_P = True except Exception as exc: MIX_P = False 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') # dataset settings parser.add_argument('--data_path', metavar='DIR', help='path to dataset') parser.add_argument('--info_path', metavar='DIR', help='path to dataset info') parser.add_argument('--eval_path', metavar='DIR', help='path to evaluation data') parser.add_argument('--eval_name', type=str, default='market', help='evaluation data name') # training settings parser.add_argument('-a', '--arch', metavar='ARCH', default='resnet50', choices=model_names, help='model arch: ' + '|'.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=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('-p', '--print-freq', default=10, type=int, metavar='N', help='print frequency (default: 10)') parser.add_argument('--save_freq', default=10, type=int, metavar='N', help='save checkpoint frequency (default: 10)') parser.add_argument('--eval_freq', default=10, type=int, metavar='N', help='evaluation frequency (default: 10)') parser.add_argument('--resume', default='', type=str, metavar='PATH', help='path to latest checkpoint (default: none)') parser.add_argument('--auto_resume', type=int, default=1) parser.add_argument('--mix', type=int, default=1) parser.add_argument('--workdir', default="", help="The working directory.") parser.add_argument('--snap_dir', default='snapshots', type=str) parser.add_argument('--log_dir', default='logs', type=str) parser.add_argument('--eval_only', default=0, type=int, help='if only do evaluate') # optimizer settings parser.add_argument('--optimizer', default='LARS', type=str, help='one of LARS|SGD') parser.add_argument('--warmup_epochs', default=0, type=int, metavar='N', help='number of epochs to warmup learning rate (default: 10)') parser.add_argument('--lr', '--learning-rate', default=0.3, type=float, metavar='LR', help='initial learning rate', dest='lr') parser.add_argument('--schedule', default='60,80', type=str, help='learning rate schedule (when to drop lr by 10x)') 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') # distributed settings 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', type=int, default=1, 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') # SimCLR settings parser.add_argument('--moco_dim', default=128, type=int, help='feature dimension (default: 128)') parser.add_argument('--T', default=0.1, type=float, help='softmax temperature (default: 0.1)') parser.add_argument('--moco_k', default=65536, type=int, help='queue size; number of negative keys (default: 65536)') parser.add_argument('--moco_m', default=0.999, type=float, help='moco momentum of updating key encoder (default: 0.999)') parser.add_argument('--mlp', type=int, default=1, help='use mlp head') parser.add_argument('--cos', default=1, type=int, help='use cosine lr schedule') parser.add_argument('--aug_type', default='ori', type=str) parser.add_argument('--mean_type', default='lup', type=str) def main(): args = parser.parse_args() args.auto_resume = True if args.auto_resume >= 1 else False args.mix = True if args.mix >= 1 else False args.eval_only = True if args.eval_only >= 1 else False args.cos = True if args.cos >= 1 else False args.mlp = True if args.mlp >= 1 else False args.schedule = [int(x) for x in args.schedule.split(',')] if args.workdir: args.snap_dir = os.path.join(args.workdir, args.snap_dir) args.log_dir = os.path.join(args.workdir, args.log_dir) args.data_path = os.path.join(args.workdir, args.data_path) args.info_path = os.path.join(args.workdir, args.info_path) args.eval_path = os.path.join(args.workdir, args.eval_path) cudnn.benchmark = True if args.seed is not None: random.seed(args.seed) torch.manual_seed(args.seed) torch.cuda.manual_seed(args.seed) np.random.seed(args.seed) cudnn.deterministic = True cudnn.benchmark = False 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 args.mix = MIX_P and args.mix 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 else: if not os.path.exists(args.snap_dir): os.makedirs(args.snap_dir) if not os.path.exists(args.log_dir): os.makedirs(args.log_dir) txt_log_name = "train_log_" + time.strftime("%Y-%m-%d_%H-%M-%S", time.gmtime()) + '.txt' sys.stdout = Logger(os.path.join(args.log_dir, txt_log_name)) 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) if args.rank == 0: print(f"Training configs:") print(f"{args} \n") # Data loading code train_transformer = get_lup_transformer(args.aug_type, args.mean_type) val_transformer = get_reid_test_transformer(args.mean_type) if args.rank == 0: print(f"Training Transformers: \n {str(train_transformer)}") print(f"Testing Transformers: \n {str(val_transformer)} \n") train_dataset = PersonDataset(args.data_path, args.info_path, train_transformer) val_dataset = CMDM(args.eval_path, args.eval_name, 'test', val_transformer) if args.rank == 0: print(f"Training set: {train_dataset}") print(f"Testing set: {val_dataset}") # create model if args.rank == 0: print("Creating model '{}'".format(args.arch)) model = MoCo(models.__dict__[args.arch], dim=args.moco_dim, K=args.moco_k, m=args.moco_m, T=args.T, mlp=args.mlp) # model = nn.SyncBatchNorm.convert_sync_batchnorm(model) 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 if args.optimizer == 'LARS': args.lr_mult = args.batch_size / 256 else: args.lr_mult = 1. 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]) model = torch.nn.parallel.DistributedDataParallel(model, device_ids=[args.gpu], broadcast_buffers=False) 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.") # define loss function (criterion) and optimizer criterion = nn.CrossEntropyLoss().cuda(args.gpu) if args.optimizer == 'LARS': if args.rank == 0: print("Use LARS as optimizer.") optimizer = get_lars_optimizer(model, args.lr_mult * args.lr, momentum=args.momentum, weight_decay=args.weight_decay) else: if args.rank == 0: print("Use SGD as optimizer.") optimizer = torch.optim.SGD(model.parameters(), args.lr, momentum=args.momentum, weight_decay=args.weight_decay) if args.auto_resume: # 'ckpt_{:04d}.pth' # if have ckpt_latest.pth, load it. Or find the record with max epoch latest_checkpoint = os.path.join(args.snap_dir, 'ckpt_latest.pth') if os.path.isfile(latest_checkpoint): args.resume = latest_checkpoint else: snaps = os.listdir(args.snap_dir) snaps = [x for x in snaps if x.endswith('.pth')] if len(snaps) > 0: max_pre_epoch = 0 for snap in snaps: tmp_epoch = int(snap[:-4].split('_')[1]) if tmp_epoch > max_pre_epoch: max_pre_epoch = tmp_epoch latest_checkpoint = 'ckpt_{:04d}.pth'.format(max_pre_epoch) args.resume = os.path.join(args.snap_dir, latest_checkpoint) else: print(f'=>rank[{args.rank}] no previous snapshots eixst for auto-resume.') if args.resume: if os.path.isfile(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("=>rank[{}] loaded checkpoint '{}' (epoch {})" .format(args.rank, args.resume, checkpoint['epoch'])) else: print("=>rank[{}] no checkpoint found at '{}'".format(args.rank, args.resume)) if args.distributed: train_sampler = torch.utils.data.distributed.DistributedSampler(train_dataset) val_sampler = InferenceSampler(len(val_dataset)) else: train_sampler = None val_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 = torch.utils.data.DataLoader(val_dataset, batch_size=128, pin_memory=True, num_workers=args.workers, sampler=val_sampler, drop_last=False) if args.eval_only: mAP, cmc = evaluate(val_loader, model, args) if args.rank == 0: print_scores(mAP, cmc) return for epoch in range(args.start_epoch, args.epochs): if not args.multiprocessing_distributed or (args.multiprocessing_distributed and args.rank == 0): print("PROGRESS: {:02d}%".format(int(epoch / args.epochs * 100))) if args.distributed: train_sampler.set_epoch(epoch) adjust_learning_rate(optimizer, epoch, args) # train for one epoch epoch_start = time.time() loss, acc1 = train(train_loader, model, criterion, optimizer, epoch, args) is_save = args.multiprocessing_distributed and args.rank == 0 and args.save_freq > 0 is_save |= not args.multiprocessing_distributed time_str = time.strftime("%Y/%m/%d-%H:%M:%S", time.gmtime()) if args.rank == 0: print(f"{time_str}: Epoch[{epoch}]; Loss={loss:.3f}; Acc={acc1:.3f}; " + f"Time={(time.time() - epoch_start):.3f}; " + f"LR={optimizer.param_groups[0]['lr']:.2e}") if args.eval_freq > 0 and (epoch % args.eval_freq == 0 or epoch == args.epochs - 1): eval_time = time.time() if args.rank == 0: print(f"Evaluating Epoch[{epoch}] ...") mAP, cmc = evaluate(val_loader, model, args) if args.rank == 0: print_scores(mAP, cmc, p_str=f"Epoch[{epoch}]: ") print(f'Evaluation using {time.time() - eval_time:.2f}s.') if is_save: latest = os.path.join(args.snap_dir, 'ckpt_latest.pth') filename = os.path.join(args.snap_dir, 'ckpt_{:04d}.pth'.format(epoch)) save_cur = epoch % args.save_freq == 0 or epoch == args.epochs - 1 print(f'Saving checkpoint to {filename if save_cur else latest} ', end='') save_time = time.time() save_checkpoint({'epoch': epoch + 1, 'arch': args.arch, 'state_dict': model.state_dict(), 'optimizer' : optimizer.state_dict()}, save_cur=save_cur, cur_name=filename, lastest_name=latest) print(f'Using {time.time() - save_time:.2f}s.') def train(train_loader, model, criterion, optimizer, epoch, args): batch_time = AverageMeter('T', ':.3f') data_time = AverageMeter('DT', ':.3f') losses = AverageMeter('Loss', ':.4f') top1 = AverageMeter('Acc@1', ':.2f') top5 = AverageMeter('Acc@5', ':.2f') progress = ProgressMeter( len(train_loader), [batch_time, data_time, losses, top1, top5], prefix="Epoch: [{}]".format(epoch)) if args.print_freq > 1: print_freq = (len(train_loader) + args.print_freq - 1) // args.print_freq else: print_freq = - args.print_freq // 1 print_freq = max(print_freq, 1) # switch to train mode model.train() end = time.time() scaler = GradScaler() if args.mix else None 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 if args.mix: with autocast(): output, target = model(im_q=images[0], im_k=images[1]) loss = criterion(output, target) else: output, target = model(im_q=images[0], im_k=images[1]) loss = criterion(output, target) # acc1/acc5 are (K+1)-way contrast classifier accuracy # measure accuracy and record loss acc1, acc5 = accuracy(output, target, topk=(1, 5)) losses.update(loss.item(), images[0].size(0)) top1.update(acc1[0], images[0].size(0)) top5.update(acc5[0], images[0].size(0)) # compute gradient and do SGD step optimizer.zero_grad() if args.mix: scaler.scale(loss).backward() scaler.step(optimizer) scaler.update() else: loss.backward() optimizer.step() # measure elapsed time batch_time.update(time.time() - end) end = time.time() if args.rank == 0 and i % print_freq == 0: progress.display(i) return losses.avg, top1.avg def evaluate(val_loader, model, args): model.eval() feats, pids, cams, flags = [], [], [], [] with torch.no_grad(): for i, data in enumerate(val_loader): img, pid, cam, flag = data[0:4] feat = model(img.cuda(args.gpu, non_blocking=True)) feats.append(feat) pids.append(pid.cuda(args.gpu, non_blocking=True)) cams.append(cam.cuda(args.gpu, non_blocking=True)) flags.append(flag.cuda(args.gpu, non_blocking=True)) feats = torch.cat(feats, dim=0) pids = torch.cat(pids, dim=0) cams = torch.cat(cams, dim=0) flags = torch.cat(flags, dim=0) dist.barrier() feats = gather_tensors(feats) pids = gather_tensors(pids) cams = gather_tensors(cams) flags = gather_tensors(flags) if args.rank == 0: q_idx = flags > 0 g_idx = flags < 1 mAP, cmc = compute_reid_score(feats[q_idx], pids[q_idx], cams[q_idx], feats[g_idx], pids[g_idx], cams[g_idx]) return mAP, cmc else: return 0, 0 def save_checkpoint(state, save_cur, cur_name='ckpt.pth', lastest_name='latest.pth'): torch.save(state, lastest_name) if save_cur: shutil.copyfile(lastest_name, cur_name) 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 ProgressMeter(object): def __init__(self, iters_per_epoch, meters, iters=None, prefix=""): self.batch_fmtstr = self._get_batch_fmtstr(iters_per_epoch) if not iters is None: self.iters_fmtstr = self._get_batch_fmtstr(iters) else: self.iters_fmtstr = None self.iters_per_epoch = iters_per_epoch self.iters = iters self.meters = meters self.prefix = prefix def display(self, i): entries = self.prefix + self.batch_fmtstr.format(i % self.iters_per_epoch) if not self.iters_fmtstr is None: entries += self.iters_fmtstr.format(i) entries = [entries] entries += [str(meter) for meter in self.meters] print(' '.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) + ']' class Logger(object): def __init__(self, fpath=None): self.console = sys.stdout self.file = None if fpath is not None: log_dir = os.path.dirname(fpath) if not os.path.exists(log_dir): os.makedirs(log_dir) self.file = open(fpath, 'w') def __del__(self): self.close() def __enter__(self): pass def __exit__(self, *args): self.close() def write(self, msg): self.console.write(msg) if self.file is not None: self.file.write(msg) def flush(self): self.console.flush() if self.file is not None: self.file.flush() os.fsync(self.file.fileno()) def close(self): self.console.close() if self.file is not None: self.file.close() 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 elif 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 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].view(-1).float().sum(0, keepdim=True) res.append(correct_k.mul_(100.0 / batch_size)) return res if __name__ == '__main__': main()