import argparse import os import random import shutil import time import warnings 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 segmentation_dataset import SegmentationDataset, VAL_PARTITION, TRAIN_PARTITION import numpy as np # Uncomment the expected model below # ViT from ViT.ViT import vit_base_patch16_224 as vit # from ViT.ViT import vit_large_patch16_224 as vit # ViT-AugReg # from ViT.ViT_new import vit_small_patch16_224 as vit # from ViT.ViT_new import vit_base_patch16_224 as vit # from ViT.ViT_new import vit_large_patch16_224 as vit # DeiT # from ViT.ViT import deit_base_patch16_224 as vit # from ViT.ViT import deit_small_patch16_224 as vit from ViT.explainer import generate_relevance, get_image_with_relevance import torchvision import cv2 from torch.utils.tensorboard import SummaryWriter import json model_names = sorted(name for name in models.__dict__ if name.islower() and not name.startswith("__") and callable(models.__dict__[name])) model_names.append("vit") parser = argparse.ArgumentParser(description='PyTorch ImageNet Training') parser.add_argument('--data', metavar='DATA', help='path to dataset') parser.add_argument('--seg_data', metavar='SEG_DATA', help='path to segmentation dataset') parser.add_argument('-j', '--workers', default=4, type=int, metavar='N', help='number of data loading workers (default: 4)') parser.add_argument('--epochs', default=50, 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=8, 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=3e-6, type=float, metavar='LR', help='initial learning rate', dest='lr') 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=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('-e', '--evaluate', dest='evaluate', action='store_true', help='evaluate model on validation set') parser.add_argument('--pretrained', dest='pretrained', action='store_true', help='use pre-trained model') 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('--save_interval', default=20, type=int, help='interval to save segmentation results.') parser.add_argument('--num_samples', default=3, type=int, help='number of samples per class for training') 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') parser.add_argument('--lambda_seg', default=0.8, type=float, help='influence of segmentation loss.') parser.add_argument('--lambda_acc', default=0.2, type=float, help='influence of accuracy loss.') parser.add_argument('--experiment_folder', default=None, type=str, help='path to folder to use for experiment.') parser.add_argument('--num_classes', default=500, type=int, help='coefficient of loss for segmentation foreground.') parser.add_argument('--temperature', default=1, type=float, help='temperature for softmax (mostly for DeiT).') best_loss = float('inf') def main(): args = parser.parse_args() if args.experiment_folder is None: args.experiment_folder = f'experiment/' \ f'lr_{args.lr}_seg_{args.lambda_seg}_acc_{args.lambda_acc}' if args.temperature != 1: args.experiment_folder = args.experiment_folder + f'_tempera_{args.temperature}' if args.batch_size != 10: args.experiment_folder = args.experiment_folder + f'_bs_{args.batch_size}' if args.num_classes != 500: args.experiment_folder = args.experiment_folder + f'_num_classes_{args.num_classes}' if args.num_samples != 3: args.experiment_folder = args.experiment_folder + f'_num_samples_{args.num_samples}' if args.epochs != 150: args.experiment_folder = args.experiment_folder + f'_num_epochs_{args.epochs}' if os.path.exists(args.experiment_folder): raise Exception(f"Experiment path {args.experiment_folder} already exists!") os.mkdir(args.experiment_folder) os.mkdir(f'{args.experiment_folder}/train_samples') os.mkdir(f'{args.experiment_folder}/val_samples') with open(f'{args.experiment_folder}/commandline_args.txt', 'w') as f: json.dump(args.__dict__, f, indent=2) 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): global best_loss args.gpu = gpu 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") model = vit(pretrained=True).cuda() model.train() print("done") if not torch.cuda.is_available(): print('using CPU, this will be slow') elif 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: # DataParallel will divide and allocate batch_size to all available GPUs print("start") model = torch.nn.DataParallel(model).cuda() # define loss function (criterion) and optimizer criterion = nn.CrossEntropyLoss().cuda(args.gpu) optimizer = torch.optim.AdamW(model.parameters(), args.lr, 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'] best_loss = checkpoint['best_loss'] if args.gpu is not None: # best_loss may be from a checkpoint from a different GPU best_loss = best_loss.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 train_dataset = SegmentationDataset(args.seg_data, args.data, partition=TRAIN_PARTITION, train_classes=args.num_classes, num_samples=args.num_samples) 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=False, num_workers=args.workers, pin_memory=True, sampler=train_sampler) val_dataset = SegmentationDataset(args.seg_data, args.data, partition=VAL_PARTITION, train_classes=args.num_classes, num_samples=1) val_loader = torch.utils.data.DataLoader( val_dataset, batch_size=5, shuffle=False, num_workers=args.workers, pin_memory=True) if args.evaluate: validate(val_loader, model, criterion, 0, args) return for epoch in range(args.start_epoch, args.epochs): if args.distributed: train_sampler.set_epoch(epoch) adjust_learning_rate(optimizer, epoch, args) log_dir = os.path.join(args.experiment_folder, 'logs') logger = SummaryWriter(log_dir=log_dir) args.logger = logger # train for one epoch train(train_loader, model, criterion, optimizer, epoch, args) # evaluate on validation set loss1 = validate(val_loader, model, criterion, epoch, args) # remember best acc@1 and save checkpoint is_best = loss1 < best_loss best_loss = min(loss1, best_loss) if not args.multiprocessing_distributed or (args.multiprocessing_distributed and args.rank % ngpus_per_node == 0): save_checkpoint({ 'epoch': epoch + 1, 'state_dict': model.state_dict(), 'best_loss': best_loss, 'optimizer' : optimizer.state_dict(), }, is_best, folder=args.experiment_folder) def train(train_loader, model, criterion, optimizer, epoch, args): mse_criterion = torch.nn.MSELoss(reduction='mean') losses = AverageMeter('Loss', ':.4e') top1 = AverageMeter('Acc@1', ':6.2f') top5 = AverageMeter('Acc@5', ':6.2f') orig_top1 = AverageMeter('Acc@1_orig', ':6.2f') orig_top5 = AverageMeter('Acc@5_orig', ':6.2f') progress = ProgressMeter( len(train_loader), [losses, top1, top5, orig_top1, orig_top5], prefix="Epoch: [{}]".format(epoch)) orig_model = vit(pretrained=True).cuda() orig_model.eval() # switch to train mode model.train() for i, (seg_map, image_ten, class_name) in enumerate(train_loader): if torch.cuda.is_available(): image_ten = image_ten.cuda(args.gpu, non_blocking=True) seg_map = seg_map.cuda(args.gpu, non_blocking=True) class_name = class_name.cuda(args.gpu, non_blocking=True) image_ten.requires_grad = True output = model(image_ten) # segmentation loss batch_size = image_ten.shape[0] index = class_name if index == None: index = np.argmax(output.cpu().data.numpy(), axis=-1) index = torch.tensor(index) one_hot = np.zeros((batch_size, output.shape[-1]), dtype=np.float32) one_hot[torch.arange(batch_size), index.data.cpu().numpy()] = 1 one_hot = torch.from_numpy(one_hot).requires_grad_(True) one_hot = torch.sum(one_hot.to(image_ten.device) * output) model.zero_grad() relevance = torch.autograd.grad(one_hot, image_ten, retain_graph=True)[0] reverse_seg_map = seg_map.clone() reverse_seg_map[reverse_seg_map == 1] = -1 reverse_seg_map[reverse_seg_map == 0] = 1 reverse_seg_map[reverse_seg_map == -1] = 0 grad_loss = mse_criterion(relevance * reverse_seg_map, torch.zeros_like(relevance)) segmentation_loss = grad_loss # classification loss with torch.no_grad(): output_orig = orig_model(image_ten) if args.temperature != 1: output = output / args.temperature classification_loss = criterion(output, class_name.flatten()) loss = args.lambda_seg * segmentation_loss + args.lambda_acc * classification_loss # debugging output if i % args.save_interval == 0: orig_relevance = generate_relevance(orig_model, image_ten, index=class_name) for j in range(image_ten.shape[0]): image = get_image_with_relevance(image_ten[j], torch.ones_like(image_ten[j])) new_vis = get_image_with_relevance(image_ten[j]*relevance[j], torch.ones_like(image_ten[j])) old_vis = get_image_with_relevance(image_ten[j], orig_relevance[j]) gt = get_image_with_relevance(image_ten[j], seg_map[j]) h_img = cv2.hconcat([image, gt, old_vis, new_vis]) cv2.imwrite(f'{args.experiment_folder}/train_samples/res_{i}_{j}.jpg', h_img) # measure accuracy and record loss acc1, acc5 = accuracy(output, class_name, topk=(1, 5)) losses.update(loss.item(), image_ten.size(0)) top1.update(acc1[0], image_ten.size(0)) top5.update(acc5[0], image_ten.size(0)) # metrics for original vit acc1_orig, acc5_orig = accuracy(output_orig, class_name, topk=(1, 5)) orig_top1.update(acc1_orig[0], image_ten.size(0)) orig_top5.update(acc5_orig[0], image_ten.size(0)) # compute gradient and do SGD step optimizer.zero_grad() loss.backward() optimizer.step() if i % args.print_freq == 0: progress.display(i) args.logger.add_scalar('{}/{}'.format('train', 'segmentation_loss'), segmentation_loss, epoch*len(train_loader)+i) args.logger.add_scalar('{}/{}'.format('train', 'classification_loss'), classification_loss, epoch * len(train_loader) + i) args.logger.add_scalar('{}/{}'.format('train', 'orig_top1'), acc1_orig, epoch * len(train_loader) + i) args.logger.add_scalar('{}/{}'.format('train', 'top1'), acc1, epoch * len(train_loader) + i) args.logger.add_scalar('{}/{}'.format('train', 'orig_top5'), acc5_orig, epoch * len(train_loader) + i) args.logger.add_scalar('{}/{}'.format('train', 'top5'), acc5, epoch * len(train_loader) + i) args.logger.add_scalar('{}/{}'.format('train', 'tot_loss'), loss, epoch * len(train_loader) + i) def validate(val_loader, model, criterion, epoch, args): mse_criterion = torch.nn.MSELoss(reduction='mean') losses = AverageMeter('Loss', ':.4e') top1 = AverageMeter('Acc@1', ':6.2f') top5 = AverageMeter('Acc@5', ':6.2f') orig_top1 = AverageMeter('Acc@1_orig', ':6.2f') orig_top5 = AverageMeter('Acc@5_orig', ':6.2f') progress = ProgressMeter( len(val_loader), [losses, top1, top5, orig_top1, orig_top5], prefix="Epoch: [{}]".format(val_loader)) # switch to evaluate mode model.eval() orig_model = vit(pretrained=True).cuda() orig_model.eval() with torch.no_grad(): for i, (seg_map, image_ten, class_name) in enumerate(val_loader): if args.gpu is not None: image_ten = image_ten.cuda(args.gpu, non_blocking=True) if torch.cuda.is_available(): seg_map = seg_map.cuda(args.gpu, non_blocking=True) class_name = class_name.cuda(args.gpu, non_blocking=True) with torch.enable_grad(): image_ten.requires_grad = True output = model(image_ten) # segmentation loss batch_size = image_ten.shape[0] index = class_name if index == None: index = np.argmax(output.cpu().data.numpy(), axis=-1) index = torch.tensor(index) one_hot = np.zeros((batch_size, output.shape[-1]), dtype=np.float32) one_hot[torch.arange(batch_size), index.data.cpu().numpy()] = 1 one_hot = torch.from_numpy(one_hot).requires_grad_(True) one_hot = torch.sum(one_hot.to(image_ten.device) * output) model.zero_grad() relevance = torch.autograd.grad(one_hot, image_ten)[0] reverse_seg_map = seg_map.clone() reverse_seg_map[reverse_seg_map == 1] = -1 reverse_seg_map[reverse_seg_map == 0] = 1 reverse_seg_map[reverse_seg_map == -1] = 0 grad_loss = mse_criterion(relevance * reverse_seg_map, torch.zeros_like(relevance)) segmentation_loss = grad_loss # classification loss output = model(image_ten) with torch.no_grad(): output_orig = orig_model(image_ten) if args.temperature != 1: output = output / args.temperature classification_loss = criterion(output, class_name.flatten()) loss = args.lambda_seg * segmentation_loss + args.lambda_acc * classification_loss # save results if i % args.save_interval == 0: with torch.enable_grad(): orig_relevance = generate_relevance(orig_model, image_ten, index=class_name) for j in range(image_ten.shape[0]): image = get_image_with_relevance(image_ten[j], torch.ones_like(image_ten[j])) new_vis = get_image_with_relevance(image_ten[j]*relevance[j], torch.ones_like(image_ten[j])) old_vis = get_image_with_relevance(image_ten[j], orig_relevance[j]) gt = get_image_with_relevance(image_ten[j], seg_map[j]) h_img = cv2.hconcat([image, gt, old_vis, new_vis]) cv2.imwrite(f'{args.experiment_folder}/val_samples/res_{i}_{j}.jpg', h_img) # measure accuracy and record loss acc1, acc5 = accuracy(output, class_name, topk=(1, 5)) losses.update(loss.item(), image_ten.size(0)) top1.update(acc1[0], image_ten.size(0)) top5.update(acc5[0], image_ten.size(0)) # metrics for original vit acc1_orig, acc5_orig = accuracy(output_orig, class_name, topk=(1, 5)) orig_top1.update(acc1_orig[0], image_ten.size(0)) orig_top5.update(acc5_orig[0], image_ten.size(0)) if i % args.print_freq == 0: progress.display(i) args.logger.add_scalar('{}/{}'.format('val', 'segmentation_loss'), segmentation_loss, epoch * len(val_loader) + i) args.logger.add_scalar('{}/{}'.format('val', 'classification_loss'), classification_loss, epoch * len(val_loader) + i) args.logger.add_scalar('{}/{}'.format('val', 'orig_top1'), acc1_orig, epoch * len(val_loader) + i) args.logger.add_scalar('{}/{}'.format('val', 'top1'), acc1, epoch * len(val_loader) + i) args.logger.add_scalar('{}/{}'.format('val', 'orig_top5'), acc5_orig, epoch * len(val_loader) + i) args.logger.add_scalar('{}/{}'.format('val', 'top5'), acc5, epoch * len(val_loader) + i) args.logger.add_scalar('{}/{}'.format('val', 'tot_loss'), loss, epoch * len(val_loader) + i) # TODO: this should also be done with the ProgressMeter print(' * Acc@1 {top1.avg:.3f} Acc@5 {top5.avg:.3f}' .format(top1=top1, top5=top5)) return losses.avg def save_checkpoint(state, is_best, folder, filename='checkpoint.pth.tar'): torch.save(state, f'{folder}/{filename}') if is_best: shutil.copyfile(f'{folder}/{filename}', f'{folder}/model_best.pth.tar') 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, 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, epoch, args): """Sets the learning rate to the initial LR decayed by 10 every 30 epochs""" lr = args.lr * (0.85 ** (epoch // 2)) 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].reshape(-1).float().sum(0, keepdim=True) res.append(correct_k.mul_(100.0 / batch_size)) return res if __name__ == '__main__': main()