Some cleanup and AFLW working

natanielruiz

2017-08-13 e306e86e925e4211c1c2d2f68de45d5e55f3e215

 code/train_AFLW.py

@@ -41,6 +41,10 @@
          default='', type=str)
    parser.add_argument('--filename_list', dest='filename_list', help='Path to text file containing relative paths for every example.',
          default='', type=str)
    parser.add_argument('--finetune', dest='finetune', help='Boolean: finetune or from Imagenet pretrain.',
          default=False, type=bool)
    parser.add_argument('--snapshot', dest='snapshot', help='Path to finetune snapshot.',
          default='', type=str)

    args = parser.parse_args()

@@ -98,14 +102,18 @@
    model = hopenet.Hopenet(torchvision.models.resnet.Bottleneck, [3, 4, 6, 3], 66)
    # ResNet18
    # model = hopenet.Hopenet(torchvision.models.resnet.BasicBlock, [2, 2, 2, 2], 66)
    load_filtered_state_dict(model, model_zoo.load_url(model_urls['resnet50']))

    if args.finetune:
        model.load_state_dict(torch.load(args.snapshot))
    else:
        load_filtered_state_dict(model, model_zoo.load_url(model_urls['resnet50']))

    print 'Loading data.'

    transformations = transforms.Compose([transforms.Scale(224),transforms.RandomCrop(224),
                                          transforms.ToTensor()])

    pose_dataset = datasets.Pose_300W_LP_binned(args.data_dir, args.filename_list,
    pose_dataset = datasets.AFLW(args.data_dir, args.filename_list,
                                transformations)
    train_loader = torch.utils.data.DataLoader(dataset=pose_dataset,
                                               batch_size=batch_size,
@@ -113,10 +121,10 @@
                                               num_workers=2)

    model.cuda(gpu)
    criterion = nn.CrossEntropyLoss().cuda()
    reg_criterion = nn.MSELoss().cuda()
    criterion = nn.CrossEntropyLoss().cuda(gpu)
    reg_criterion = nn.MSELoss().cuda(gpu)
    # Regression loss coefficient
    alpha = 0.01
    alpha = 0.1

    idx_tensor = [idx for idx in xrange(66)]
    idx_tensor = torch.FloatTensor(idx_tensor).cuda(gpu)
@@ -124,6 +132,9 @@
    optimizer = torch.optim.Adam([{'params': get_ignored_params(model), 'lr': args.lr},
                                  {'params': get_non_ignored_params(model), 'lr': args.lr * 10}],
                                  lr = args.lr)
    # optimizer = torch.optim.SGD([{'params': get_ignored_params(model), 'lr': args.lr},
    #                               {'params': get_non_ignored_params(model), 'lr': args.lr * 10}],
    #                               lr = args.lr, momentum = 0.9)

    print 'Ready to train network.'

@@ -135,7 +146,6 @@
            label_roll = Variable(labels[:,2].cuda(gpu))

            optimizer.zero_grad()
            model.zero_grad()

            yaw, pitch, roll = model(images)

@@ -175,13 +185,13 @@
                       %(epoch+1, num_epochs, i+1, len(pose_dataset)//batch_size, loss_yaw.data[0], loss_pitch.data[0], loss_roll.data[0]))
                if epoch == 0:
                    torch.save(model.state_dict(),
                    'output/snapshots/resnet50_AFW_iter_'+ str(i+1) + '.pkl')
                    'output/snapshots/resnet50_AFLW_finetuned_iter_'+ str(i+1) + '.pkl')

        # Save models at numbered epochs.
        if epoch % 1 == 0 and epoch < num_epochs - 1:
            print 'Taking snapshot...'
            torch.save(model.state_dict(),
            'output/snapshots/resnet50_AFW_epoch_'+ str(epoch+1) + '.pkl')
            'output/snapshots/resnet50_AFLW_finetuned_epoch_'+ str(epoch+1) + '.pkl')

    # Save the final Trained Model
    torch.save(model.state_dict(), 'output/snapshots/resnet50_AFLW_epoch' + str(epoch+1) + '.pkl')
    torch.save(model.state_dict(), 'output/snapshots/resnet50_AFLW_finetuned_epoch_' + str(epoch+1) + '.pkl')