mit_sceneparsing_benchmark: fix train error, support "same" config that does not resize image

[helinwang]

With this fix I could train with mit_sceneparsing_benchmark dataset.

config:

model:
    arch: fcn8s
data:
    dataset: mit_sceneparsing_benchmark
    train_split: training
    val_split: validation
    img_rows: 512
    img_cols: 512
    path: /home/helin/dataset/ADEChallengeData2016/
training:
    train_iters: 300000
    batch_size: 1
    val_interval: 1000
    n_workers: 16
    print_interval: 50
    optimizer:
        name: 'sgd'
        lr: 1.0e-10
        weight_decay: 0.0005
        momentum: 0.99
    loss:
        name: 'cross_entropy'
        size_average: False
    lr_schedule:
    resume:

[meetps]

Thanks!