img_layers.py 701 字节
Newer Older
1 2 3 4 5 6 7 8 9
from paddle.trainer_config_helpers import *

settings(
    learning_rate=1e-3,
    batch_size=1000
)

img = data_layer(name='image', size=256*256)

10 11
# the parse_conv in config_parse.py is not strictly accurate when filter_size
# is not square. So here set square filter_size.
12
img_conv = img_conv_layer(input=img, num_channels=1, num_filters=64,
13
                          filter_size=(32, 32), padding=(1, 1), stride=(1, 1),
14 15 16 17 18 19 20 21
                          act=LinearActivation())
img_bn = batch_norm_layer(input=img_conv, act=ReluActivation())

img_norm = img_cmrnorm_layer(input=img_bn, size=32)

img_pool = img_pool_layer(input=img_conv, pool_size=32, pool_type=MaxPooling())


22
outputs(img_pool, img_norm)