def mnasnet_backbone(k, e):
"""Creates a mnasnet-like model with a certain type
of MBConv layers (k, e).
"""
blocks_args = [
'r1_k3_s11_e1_i32_o16_noskip',
'r4_k' + str(k) + '_s22_e' + str(e) + '_i16_o24',
'r4_k' + str(k) + '_s22_e' + str(e) + '_i24_o40',
'r4_k' + str(k) + '_s22_e' + str(e) + '_i40_o80',
'r4_k' + str(k) + '_s11_e' + str(e) + '_i80_o96',
'r4_k' + str(k) + '_s22_e' + str(e) + '_i96_o192',
'r1_k3_s11_e6_i192_o320_noskip'
]
decoder = MnasNetDecoder()
global_params = model_def.GlobalParams(batch_norm_momentum=0.99,
batch_norm_epsilon=1e-3,
dropout_rate=0.2,
data_format='channels_last',
num_classes=1000,
depth_multiplier=None,
depth_divisor=8,
expratio=e,
kernel=k,
min_depth=None)
return decoder.decode(blocks_args), global_params
def parse_netarch(parse_lambda_dir, depth_multiplier=None):
"""Creates the RNAS found model. No need to hard-code
model, it parses the output of previous search
Args:
depth_multiplier: multiplier to number of filters per layer.
Returns:
blocks_args: a list of BlocksArgs for internal MnasNet blocks.
global_params: GlobalParams, global parameters for the model.
"""
# Loading too much data is slow...
tf_size_guidance = {
'compressedHistograms': 10,
'images': 0,
'scalars': 100,
'histograms': 1
}
indicator_values = parse_netarch.parse_indicators_single_path_nas(
parse_lambda_dir, tf_size_guidance)
network = parse_netarch.encode_single_path_nas_arch(indicator_values)
parse_netarch.print_net(network)
blocks_args = parse_netarch.mnasnet_encoder(network)
parse_netarch.print_encoded_net(blocks_args)
decoder = MnasNetDecoder()
global_params = model_def.GlobalParams(batch_norm_momentum=0.99,
batch_norm_epsilon=1e-3,
dropout_rate=0.2,
data_format='channels_last',
num_classes=1000,
depth_multiplier=depth_multiplier,
depth_divisor=8,
min_depth=None)
return decoder.decode(blocks_args), global_params
def parse_netarch_string(blocks_args, depth_multiplier=None):
decoder = MnasNetDecoder()
global_params = model_def.GlobalParams(
batch_norm_momentum=0.99,
batch_norm_epsilon=1e-3,
dropout_rate=0.2,
data_format='channels_last',
num_classes=1000,
depth_multiplier=depth_multiplier,
depth_divisor=8,
min_depth=None)
return decoder.decode(blocks_args), global_params
def mnasnet_3x3_1(depth_multiplier=None):
"""Creates a mnasnet-3x3-1.
"""
blocks_args = [
'r1_k3_s11_e1_i32_o16_noskip', 'r4_k3_s22_e1_i16_o24',
'r4_k3_s22_e1_i24_o40', 'r4_k3_s22_e1_i40_o80', 'r4_k3_s11_e1_i80_o96',
'r4_k3_s22_e1_i96_o192', 'r1_k3_s11_e6_i192_o320_noskip'
]
decoder = MnasNetDecoder()
global_params = model_def.GlobalParams(batch_norm_momentum=0.99,
batch_norm_epsilon=1e-3,
dropout_rate=0.2,
data_format='channels_last',
num_classes=1000,
depth_multiplier=depth_multiplier,
depth_divisor=8,
min_depth=None)
return decoder.decode(blocks_args), global_params