stride = 1
SHAPE = (
84, 84, k
) # height * width * channel This cannot read from file and needs to be provided here
if not predict_mode: # if train
import input_utils as IU, misc_utils as MU
expr = MU.BMU.ExprCreaterAndResumer(MODEL_DIR,
postfix='pKf_dp%.2f_k%ds%d' %
(dropout, k, stride))
expr.redirect_output_to_logfile_if_not_on("eldar-11")
else:
import all_py_files_snapshot.input_utils as IU, all_py_files_snapshot.misc_utils as MU
MU.BMU.save_GPU_mem_keras()
MU.keras_model_serialization_bug_fix()
if resume_model:
model = expr.load_weight_and_training_config_and_state()
expr.printdebug("Checkpoint found. Resuming model at %s" %
expr.dir_lasttime)
else:
###############################
# Architecture of the network #
###############################
inputs = L.Input(shape=SHAPE)
x = inputs # inputs is used by the line "Model(inputs, ... )" below
conv1 = L.Conv2D(32, (8, 8), strides=4, padding='valid')
x = conv1(x)
def __init__(self, modelfile, meanfile):
MU.keras_model_serialization_bug_fix()
self.model = K.models.load_model(
modelfile) # this var stores the Keras Model
self.mean = np.load(meanfile)