train_test.py

from __future__ import print_function

import argparse
import pickle
import time

import numpy as np
import os
import torch
import torch.backends.cudnn as cudnn
import torch.nn.init as init
import torch.optim as optim
import torch.utils.data as data
import torchvision
from torch.autograd import Variable

from data import VOCroot, COCOroot, VOC_300, VOC_512, COCO_300, COCO_512, COCO_mobile_300, AnnotationTransform, \
    COCODetection, VOCDetection, detection_collate, BaseTransform, preproc
from layers.functions import Detect, PriorBox
from layers.modules import MultiBoxLoss
from utils.timer import Timer


def str2bool(v):
    return v.lower() in ("yes", "true", "t", "1")


parser = argparse.ArgumentParser(
    description='Receptive Field Block Net Training')
parser.add_argument('-v', '--version', default='SSD_HarDNet68',
                    help='SSD_vgg | SSD_HarDNet68 | SSD_HarDNet85 | RFB_HarDNet68 | RFB_HarDNet85')
parser.add_argument('-s', '--size', default='512',
                    help='300 or 512 input size.')
parser.add_argument('-d', '--dataset', default='COCO',
                    help='VOC or COCO dataset')
parser.add_argument(
    '--basenet', default='weights/vgg16_reducedfc.pth', help='pretrained base model')
parser.add_argument('--jaccard_threshold', default=0.5,
                    type=float, help='Min Jaccard index for matching')
parser.add_argument('-b', '--batch_size', default=32,
                    type=int, help='Batch size for training')
parser.add_argument('--num_workers', default=8,
                    type=int, help='Number of workers used in dataloading')
parser.add_argument('--cuda', default=True,
                    type=bool, help='Use cuda to train model')
parser.add_argument('--ngpu', default=1, type=int, help='gpus')
parser.add_argument('--lr', '--learning-rate',
                    default=4e-3, type=float, help='initial learning rate')
parser.add_argument('--momentum', default=0.9, type=float, help='momentum')

parser.add_argument('--resume_net', default=False, help='resume net for retraining')
parser.add_argument('--resume_epoch', default=0,
                    type=int, help='resume iter for retraining')

parser.add_argument('-max', '--max_epoch', default=150,
                    type=int, help='max epoch for retraining')
parser.add_argument('--weight_decay', default=1e-4,
                    type=float, help='Weight decay for SGD')
parser.add_argument('-we', '--warm_epoch', default=0,
                    type=int, help='max epoch for retraining')
parser.add_argument('--gamma', default=0.1,
                    type=float, help='Gamma update for SGD')
parser.add_argument('--log_iters', default=True,
                    type=bool, help='Print the loss at each iteration')
parser.add_argument('--save_folder', default='weights/',
                    help='Location to save checkpoint models')
parser.add_argument('--date', default='1213')
parser.add_argument('--save_frequency', default=10)
parser.add_argument('--test',default=None, help='test pretrained model')
parser.add_argument('--retest', default=False, type=bool,
                    help='test cache results')
parser.add_argument('--test_frequency', default=10)
parser.add_argument('--visdom', default=False, type=str2bool, help='Use visdom to for loss visualization')
parser.add_argument('--send_images_to_visdom', type=str2bool, default=False,
                    help='Sample a random image from each 10th batch, send it to visdom after augmentations step')
args = parser.parse_args()

save_folder = os.path.join(args.save_folder, args.version + '_' + args.size, args.date)
if not os.path.exists(save_folder):
    os.makedirs(save_folder)
test_save_dir = os.path.join(save_folder, 'ss_predict')
if not os.path.exists(test_save_dir):
    os.makedirs(test_save_dir)

log_file_path = save_folder + '/train' + time.strftime('_%Y-%m-%d-%H-%M', time.localtime(time.time())) + '.log'
if args.dataset == 'VOC':
    train_sets = [('2007', 'trainval'), ('2012', 'trainval')]
    cfg = (VOC_300, VOC_512)[args.size == '512']
else:
    train_sets = [('2017', 'train')]
    cfg = (COCO_300, COCO_512)[args.size == '512']

if args.version == 'SSD_vgg':
    from models.SSD_vgg import build_net
    
elif args.version == 'SSD_HarDNet68':
    from models.SSD_HarDNet68 import build_net
elif args.version == 'SSD_HarDNet85':
    from models.SSD_HarDNet85 import build_net
elif args.version == 'RFB_HarDNet68':
    from models.RFB_HarDNet68 import build_net
elif args.version == 'RFB_HarDNet85':
    from models.RFB_HarDNet85 import build_net

else:
    print('Unkown version!')
rgb_std = (1, 1, 1)
rgb_means = (104, 117, 123)
img_dim = (300, 512)[args.size == '512']

p = (0.6, 0.2)[args.version == 'RFB_mobile']
num_classes = (21, 81)[args.dataset == 'COCO']
batch_size = args.batch_size
gamma = 0.1
momentum = 0.9
if args.visdom:
    import visdom

    viz = visdom.Visdom()

net = build_net(img_dim, num_classes)
print(net)
if not args.resume_net and args.test is None:
    if args.version == 'SSD_vgg':
      base_weights = torch.load(args.basenet)
      print('Loading base network...')
      net.base.load_state_dict(base_weights)


    def xavier(param):
        init.xavier_uniform(param)


    def weights_init(m):
        for key in m.state_dict():
            if key.split('.')[-1] == 'weight':
                if 'conv' in key:
                    init.kaiming_normal(m.state_dict()[key], mode='fan_out')
                if 'bn' in key:
                    m.state_dict()[key][...] = 1
            elif key.split('.')[-1] == 'bias':
                m.state_dict()[key][...] = 0


    print('Initializing weights...')
    # initialize newly added layers' weights with kaiming_normal method
    net.extras.apply(weights_init)
    net.loc.apply(weights_init)
    net.conf.apply(weights_init)
    if 'RFB' in args.version:
        net.Norm.apply(weights_init)
    if hasattr(net, 'bridge'):
        net.bridge.apply(weights_init)        

else:
    # load resume network
    resume_net_path = os.path.join(save_folder, args.version + '_' + args.dataset + '_epoches_' + \
                                   str(args.resume_epoch) + '.pth')
    if args.test is not None:
        print('Loading pretrained model for testing:', args.test)
        state_dict = torch.load(args.test)
    else:                    
        print('Loading resume network', resume_net_path)
        state_dict = torch.load(resume_net_path)
    # create new OrderedDict that does not contain `module.`
    from collections import OrderedDict

    new_state_dict = OrderedDict()
    for k, v in state_dict.items():
        head = k[:7]
        if head == 'module.':
            name = k[7:]  # remove `module.`
        else:
            name = k
        new_state_dict[name] = v
    net.load_state_dict(new_state_dict)

if args.ngpu > 0:
    net = torch.nn.DataParallel(net, device_ids=list(range(args.ngpu)))

if args.cuda:
    net.cuda()
    cudnn.benchmark = True

detector = Detect(num_classes, 0, cfg)
optimizer = optim.SGD(net.parameters(), lr=args.lr,
                      momentum=args.momentum, weight_decay=args.weight_decay)
# optimizer = optim.RMSprop(net.parameters(), lr=args.lr,alpha = 0.9, eps=1e-08,
#                      momentum=args.momentum, weight_decay=args.weight_decay)

criterion = MultiBoxLoss(num_classes, 0.5, True, 0, True, 3, 0.5, False)
priorbox = PriorBox(cfg)
priors = Variable(priorbox.forward(), volatile=True)
# dataset
print('Loading Dataset...')
if args.dataset == 'VOC':
    testset = VOCDetection(
        VOCroot, [('2007', 'test')], None, AnnotationTransform())
    train_dataset = VOCDetection(VOCroot, train_sets, preproc(
        img_dim, rgb_means, rgb_std, p), AnnotationTransform())
elif args.dataset == 'COCO':
    testset = COCODetection(
        COCOroot, [('2017', 'val')], None)
    train_dataset = COCODetection(COCOroot, train_sets, preproc(
        img_dim, rgb_means, rgb_std, p))
else:
    print('Only VOC and COCO are supported now!')
    exit()

def test():
    torch.backends.cudnn.benchmark = True
    net.eval()
    top_k = (300, 200)[args.dataset == 'COCO']

    if args.dataset == 'VOC':
        APs, mAP = test_net(test_save_dir, net, detector, args.cuda, testset,
                            BaseTransform(net.module.size, rgb_means, rgb_std, (2, 0, 1)),
                            top_k, thresh=0.03)
        APs = [str(num) for num in APs]
        mAP = str(mAP)
        print('mAP:\n' + mAP + '\n')
    else:
        test_net(test_save_dir, net, detector, args.cuda, testset,
                 BaseTransform(net.module.size, rgb_means, rgb_std, (2, 0, 1)),
                 top_k, thresh=0.02)


def train():
    net.train()
    # loss counters
    epoch = 0
    if args.resume_net:
        epoch = 0 + args.resume_epoch
    epoch_size = len(train_dataset) // args.batch_size
    max_iter = args.max_epoch * epoch_size

    stepvalues_VOC = (150 * epoch_size, 200 * epoch_size, 250 * epoch_size)
    stepvalues_COCO = (90 * epoch_size, 120 * epoch_size, 140 * epoch_size)
    stepvalues = (stepvalues_VOC, stepvalues_COCO)[args.dataset == 'COCO']
    print('Training', args.version, 'on', train_dataset.name)
    step_index = 0

    if args.visdom:
        # initialize visdom loss plot
        lot = viz.line(
            X=torch.zeros((1,)).cpu(),
            Y=torch.zeros((1, 3)).cpu(),
            opts=dict(
                xlabel='Iteration',
                ylabel='Loss',
                title='Current SSD Training Loss',
                legend=['Loc Loss', 'Conf Loss', 'Loss']
            )
        )
        epoch_lot = viz.line(
            X=torch.zeros((1,)).cpu(),
            Y=torch.zeros((1, 3)).cpu(),
            opts=dict(
                xlabel='Epoch',
                ylabel='Loss',
                title='Epoch SSD Training Loss',
                legend=['Loc Loss', 'Conf Loss', 'Loss']
            )
        )
    if args.resume_epoch > 0:
        start_iter = args.resume_epoch * epoch_size
    else:
        start_iter = 0

    log_file = open(log_file_path, 'w')
    batch_iterator = None
    mean_loss_c = 0
    mean_loss_l = 0
    for iteration in range(start_iter, max_iter + 10):
        if (iteration % epoch_size == 0):
            # create batch iterator
            batch_iterator = iter(data.DataLoader(train_dataset, batch_size,
                                                  shuffle=True, num_workers=args.num_workers,
                                                  collate_fn=detection_collate))
            loc_loss = 0
            conf_loss = 0
            if epoch % args.save_frequency == 0 and epoch > 0:
                torch.save(net.state_dict(), os.path.join(save_folder, args.version + '_' + args.dataset + '_epoches_' +
                                                          repr(epoch) + '.pth'))
            if epoch % args.test_frequency == 0 and epoch > 0:
                net.eval()
                top_k = (300, 200)[args.dataset == 'COCO']
                if args.dataset == 'VOC':
                    APs, mAP = test_net(test_save_dir, net, detector, args.cuda, testset,
                                        BaseTransform(net.module.size, rgb_means, rgb_std, (2, 0, 1)),
                                        top_k, thresh=0.01)
                    APs = [str(num) for num in APs]
                    mAP = str(mAP)
                    log_file.write(str(iteration) + ' APs:\n' + '\n'.join(APs))
                    log_file.write('mAP:\n' + mAP + '\n')
                else:
                    test_net(test_save_dir, net, detector, args.cuda, testset,
                             BaseTransform(net.module.size, rgb_means, rgb_std, (2, 0, 1)),
                             top_k, thresh=0.01)

                net.train()
            epoch += 1

        load_t0 = time.time()
        if iteration in stepvalues:
            step_index = stepvalues.index(iteration) + 1
            if args.visdom:
                viz.line(
                    X=torch.ones((1, 3)).cpu() * epoch,
                    Y=torch.Tensor([mean_loss_l, mean_loss_c,
                                    mean_loss_l + mean_loss_c]).unsqueeze(0).cpu() / epoch_size,
                    win=epoch_lot,
                    update='append'
                )
        lr = adjust_learning_rate(optimizer, args.gamma, epoch, step_index, iteration, epoch_size)

        # load train data
        images, targets = next(batch_iterator)

        # print(np.sum([torch.sum(anno[:,-1] == 2) for anno in targets]))

        if args.cuda:
            images = Variable(images.cuda())
            targets = [Variable(anno.cuda(), volatile=True) for anno in targets]
        else:
            images = Variable(images)
            targets = [Variable(anno, volatile=True) for anno in targets]
        # forward
        out = net(images)
        # backprop
        optimizer.zero_grad()
        # arm branch loss
        loss_l, loss_c = criterion(out, priors, targets)
        # odm branch loss

        mean_loss_c += loss_c.data
        mean_loss_l += loss_l.data

        loss = loss_l + loss_c
        loss.backward()
        optimizer.step()
        load_t1 = time.time()
        if iteration % 10 == 0:
            print('Epoch:' + repr(epoch) + ' || epochiter: ' + repr(iteration % epoch_size) + '/' + repr(epoch_size)
                  + '|| Totel iter ' +
                  repr(iteration) + ' || L: %.4f C: %.4f||' % (
                      mean_loss_l / 10, mean_loss_c / 10) +
                  'Batch time: %.4f sec. ||' % (load_t1 - load_t0) + 'LR: %.8f' % (lr))
            log_file.write(
                'Epoch:' + repr(epoch) + ' || epochiter: ' + repr(iteration % epoch_size) + '/' + repr(epoch_size)
                + '|| Totel iter ' +
                repr(iteration) + ' || L: %.4f C: %.4f||' % (
                    mean_loss_l / 10, mean_loss_c / 10) +
                'Batch time: %.4f sec. ||' % (load_t1 - load_t0) + 'LR: %.8f' % (lr) + '\n')

            mean_loss_c = 0
            mean_loss_l = 0
            if args.visdom and args.send_images_to_visdom:
                random_batch_index = np.random.randint(images.size(0))
                viz.image(images.data[random_batch_index].cpu().numpy())
    log_file.close()
    torch.save(net.state_dict(), os.path.join(save_folder,
                                              'Final_' + args.version + '_' + args.dataset + '.pth'))


def adjust_learning_rate(optimizer, gamma, epoch, step_index, iteration, epoch_size):
    """Sets the learning rate
    # Adapted from PyTorch Imagenet example:
    # https://github.com/pytorch/examples/blob/master/imagenet/main.py
    """
    if epoch < args.warm_epoch:
        lr = 1e-6 + (args.lr - 1e-6) * iteration / (epoch_size * args.warm_epoch)
    else:
        lr = args.lr * (gamma ** (step_index))
    for param_group in optimizer.param_groups:
        param_group['lr'] = lr
    return lr


def test_net(save_folder, net, detector, cuda, testset, transform, max_per_image=300, thresh=0.005):
    if not os.path.exists(save_folder):
        os.mkdir(save_folder)
    # dump predictions and assoc. ground truth to text file for now
    num_images = len(testset)
    num_classes = (21, 81)[args.dataset == 'COCO']
    all_boxes = [[[] for _ in range(num_images)]
                 for _ in range(num_classes)]
    
    _t = {'im_detect': Timer(), 'misc': Timer(), 'overall': Timer()}
    det_file = os.path.join(save_folder, 'detections.pkl')

    if args.retest:
        f = open(det_file, 'rb')
        all_boxes = pickle.load(f)
        print('Evaluating detections')
        testset.evaluate_detections(all_boxes, save_folder)
        return

    for i in range(num_images):
        img = testset.pull_image(i)
        x = Variable(transform(img).unsqueeze(0), volatile=True)
        if cuda:
            x = x.cuda()

        torch.cuda.synchronize()
        _t['overall'].tic()
        _t['im_detect'].tic()
        out = net(x=x, test=True)  # forward pass
        boxes, scores = detector.forward(out, priors)
        torch.cuda.synchronize()
        detect_time = _t['im_detect'].toc()

        _t['misc'].tic()
        boxes = boxes[0]
        scores = scores[0]

        # scale each detection back up to the image
        scale = torch.Tensor([img.shape[1], img.shape[0],
                              img.shape[1], img.shape[0]]).cuda()
        boxes *= scale
        
        # Filter out bboxes with really high prob to be a background
        background_th = 0.9
        scores_class = scores.max(1)[1]
        inds = torch.where((scores_class > 0) | (scores[:,0]<background_th))[0]
        boxes = boxes[inds]
        scores = scores[inds]
        
        for j in range(1, num_classes):
            inds = torch.where(scores[:, j] > thresh)[0]
            if len(inds) == 0:
                all_boxes[j][i] = torch.empty([0, 5], dtype=torch.float32)
                continue
            c_bboxes = boxes[inds]
            c_scores = scores[inds, j]

            keep = torchvision.ops.nms(c_bboxes, c_scores.view(-1), 0.45)

            keep = keep[:50]
            c_dets = torch.cat([c_bboxes, c_scores.view(-1,1)],1)
            c_dets = c_dets[keep, :]
            all_boxes[j][i] = c_dets

        if max_per_image > 0:
            image_scores = torch.cat([all_boxes[j][i][:, -1] for j in range(1, num_classes)])
            if len(image_scores) > max_per_image:
                image_thresh = torch.sort(image_scores)[0][-max_per_image]
                for j in range(1, num_classes):
                    keep = torch.where(all_boxes[j][i][:, -1] >= image_thresh)[0]
                    all_boxes[j][i] = all_boxes[j][i][keep, :].cpu().numpy()
            else:
                for j in range(1, num_classes):
                    all_boxes[j][i] = all_boxes[j][i].cpu().numpy()

        torch.cuda.synchronize()
        nms_time = _t['misc'].toc()
        overall = _t['overall'].toc()

        if i % 20 == 0:
            print('im_detect: {:d}/{:d}  Detection: {:5.2f}ms,   NMS: {:5.2f}ms,   All: {:4.1f} fps'
                  .format(i + 1, num_images, detect_time*1000.0, nms_time*1000.0, 1/(detect_time+nms_time)))
            _t['im_detect'].clear()
            _t['misc'].clear()
        if i == 0:
            _t['overall'].clear() #discount the first inference

    print('Overall: %5.3f fps (%d images)'%(1/overall, _t['overall'].calls))

    with open(det_file, 'wb') as f:
        pickle.dump(all_boxes, f, pickle.HIGHEST_PROTOCOL)

    print('Evaluating detections')
    if args.dataset == 'VOC':
        APs, mAP = testset.evaluate_detections(all_boxes, save_folder)
        return APs, mAP
    else:
        testset.evaluate_detections(all_boxes, save_folder)


if __name__ == '__main__':
    if args.test is not None:
        test()
    else:
        train()