import os
import numpy as np
from tqdm import tqdm
from keras.models import Model
from inception_resnet_v2 import InceptionResNetV2
WEIGHTS_DIR = './weights'
MODEL_DIR = './models'
OUTPUT_WEIGHT_FILENAME = 'inception_resnet_v2_weights_tf_dim_ordering_tf_kernels.h5'
OUTPUT_WEIGHT_FILENAME_NOTOP = 'inception_resnet_v2_weights_tf_dim_ordering_tf_kernels_notop.h5'
print('Instantiating an empty InceptionResNetV2 model...')
model = InceptionResNetV2(weights=None, input_shape=(299, 299, 3))
print('Loading weights from', WEIGHTS_DIR)
for layer in tqdm(model.layers):
if layer.weights:
weights = []
for w in layer.weights:
weight_name = os.path.basename(w.name).replace(':0', '')
weight_file = layer.name + '_' + weight_name + '.npy'
weight_arr = np.load(os.path.join(WEIGHTS_DIR, weight_file))
# remove the "background class"
if weight_file.startswith('Logits_bias'):
weight_arr = weight_arr[1:]
elif weight_file.startswith('Logits_kernel'):
weight_arr = weight_arr[:, 1:]