"_data/mean_shape.pkl"
w, h = 50, 50
bx, dx = 0.5, 0.001
border_x = bx
tags_oc = None
with open(path_test, 'r') as f:
l_samples = pkl.load(f)
input = T.fmatrix("x_input")
output = T.fmatrix("y_output")
ds = FaceDataset()
x, y, l_infos = ds.sample_from_list_to_test(l_samples, w, h)
ts_batch_size = 1000
x, y, l_infos, tags_oc = artificiel_occlusion(x, y, l_infos)
list_minibatchs_vl = split_data_to_minibatchs_eval(
{"x": x, "y": y}, ts_batch_size)
fold_exp = "../../exps/" + sys.argv[1]
with open(fold_exp+"/model.pkl", 'r') as f:
stuff = pkl.load(f)
layers_infos, params_vl = stuff["layers_infos"], stuff["params_vl"]
print layers_infos
tag = stuff["tag"]
dropout = stuff["dropout"]
rng = np.random.RandomState(23455)
input = T.fmatrix("x_input")
for l in layers_infos:
l["W"], l["b"], l["rng"] = None, None, rng
model = ModelMLP(layers_infos, input, dropout=dropout)
model.set_params_vals(fold_exp+"/model.pkl")
eval_fn = get_eval_fn(model)
id_deep_conv_ae = id_deep_conv_ae.replace("(", "")
id_deep_conv_ae = id_deep_conv_ae.replace(")", "")
print len(id_deep_conv_ae), id_deep_conv_ae
path_ini_params_deep_conv_ae = init_w_deep_conv_ae_path +\
"deep_conv_ae_init_" + id_deep_conv_ae + ".pkl"
if not os.path.isfile(path_ini_params_deep_conv_ae):
deep_conv_ae.save_params(path_ini_params_deep_conv_ae)
else:
deep_conv_ae.set_params_vals(path_ini_params_deep_conv_ae)
with open(path_valid, 'r') as f:
l_samples_vl = pkl.load(f)
# convert to 3D
nbr_xx = l_samples_vl["x"].shape[0]
l_samples_vl["x"] = l_samples_vl["x"].reshape((nbr_xx, 1, h, w))
list_minibatchs_vl = split_data_to_minibatchs_eval(l_samples_vl, vl_batch_size)
max_epochs = int(1000)
lr_vl = 1e-4
lr = sharedX_value(lr_vl, name="lr")
optimizer = "momentum"
if optimizer == "adadelta":
updater = AdaDelta(decay=0.95)
elif optimizer == "momentum":
updater = Momentum(0.9,
nesterov_momentum=True,
imagenet=False,
imagenetDecay=5e-4,
max_colm_norm=False)
else:
raise ValueError("Optimizer not recognized.")
# number of the hidden layer just before the output ae. Default: None
id_code = None
model = ModelMLP(layers, input, l1_reg=0., l2_reg=0., reg_bias=False,
dropout=dropout, id_code=id_code)
aes_in = []
aes_out = []
if id_code is not None:
assert aes_out != []
# Train
# Data
tr_batch_size = 10
vl_batch_size = 8000
with open(path_valid, 'r') as f:
l_samples_vl = pkl.load(f)
list_minibatchs_vl = split_data_to_minibatchs_eval(
l_samples_vl, vl_batch_size)
max_epochs = int(1000)
lr_vl = 1e-3
lr = sharedX_value(lr_vl, name="lr")
# cost weights
separate = True
l_in = [sharedX_value(0., name="l_in"), sharedX_value(0.0, name="l_in2")]
l_out = [sharedX_value(0., name="l_out")]
l_sup = sharedX_value(1., name="l_sup")
l_code = sharedX_value(0.0, name="l_code")
if not separate:
assert l_sup.get_value() + l_in.get_value() + l_out.get_value() == 1.
if l_in[0].get_value() != 0. and aes_in == []:
raise ValueError("You setup the l_in but no aes in found.")
if l_out[0].get_value() != 0. and aes_out == []:
raise ValueError("You setup the l_out but no aes out found.")
# x, y, l_infos = ds.sample_from_list_to_test(l_samples, w, h)
ts_batch_size = 1000
with open("../../inout/data/face/" + faceset + "_data/test.pkl",
'r') as fx:
dumped = pkl.load(fx)
with open("../../inout/data/face/" + faceset + "_data/valid.pkl",
'r') as fx:
dumped_vl = pkl.load(fx)
x = dumped["x"]
y = dumped["y"]
nfids = dumped["nfids"]
bboxesT = dumped["bboxesT"]
bboxesT_original = dumped["bboxesT_original"]
base_name = dumped["base_name"]
list_minibatchs_vl = split_data_to_minibatchs_eval({
"x": x,
"y": y
}, ts_batch_size)
fold_exp = "../../exps/" + sys.argv[1]
with open(fold_exp + "/model.pkl", 'r') as f:
stuff = pkl.load(f)
layers_infos, params_vl = stuff["layers_infos"], stuff["params_vl"]
print layers_infos
tag = stuff["tag"]
dropout = stuff["dropout"]
rng = np.random.RandomState(23455)
input = T.fmatrix("x_input")
for l in layers_infos:
l["W"], l["b"], l["rng"] = None, None, rng
model = ModelMLP(layers_infos, input, dropout=dropout)
model.set_params_vals(fold_exp + "/model.pkl")
