Last active
November 10, 2017 00:07
-
-
Save tawnkramer/369fc8cf3957979827346b26b8938274 to your computer and use it in GitHub Desktop.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
def imu_rnn_lstm(seq_length, num_outputs, imu_vec_size=6): | |
from keras.layers import Input, Dense | |
from keras.models import Model | |
from keras.layers import Convolution2D, MaxPooling2D, Reshape, BatchNormalization | |
from keras.layers import Activation, Dropout, Flatten, Cropping2D, Lambda | |
from keras.layers.merge import concatenate | |
from keras.layers import LSTM | |
img_in = Input(batch_shape=(seq_length, 120,160,3), name='img_in') | |
imu_in = Input(batch_shape=(seq_length, imu_vec_size), name="imu_in") | |
x = img_in | |
x = Cropping2D(cropping=((60,0), (0,0)))(x) #trim 60 pixels off top | |
x = Convolution2D(24, (5,5), strides=(2,2), activation='relu')(x) | |
x = Convolution2D(32, (5,5), strides=(2,2), activation='relu')(x) | |
x = Convolution2D(32, (3,3), strides=(2,2), activation='relu')(x) | |
x = Convolution2D(32, (3,3), strides=(1,1), activation='relu')(x) | |
x = MaxPooling2D(pool_size=(2, 2))(x) | |
x = Flatten(name='flattened')(x) | |
x = Dense(100, activation='relu')(x) | |
x = Dropout(.1)(x) | |
y = imu_in | |
y = Dense(14, activation='relu')(y) | |
y = Dense(14, activation='relu')(y) | |
y = Dense(14, activation='relu')(y) | |
z = concatenate([x, y]) | |
z = Dense(256, activation='relu')(z) | |
z = Reshape((256, 1))(z) | |
z = LSTM(128, return_sequences=True, name="LSTM_seq")(z) | |
z = Dropout(.1)(z) | |
z = LSTM(128, return_sequences=False, name="LSTM_out")(z) | |
z = Dropout(.1)(z) | |
z = Dense(50, activation='relu')(z) | |
z = Dropout(.1)(z) | |
outputs = [] | |
for i in range(num_outputs): | |
outputs.append(Dense(1, activation='linear', name='n_outputs' + str(i))(z)) | |
model = Model(inputs=[img_in, imu_in], outputs=outputs) | |
model.compile(optimizer='adam', | |
loss='mse') | |
return model | |
def generator(data, batch_size=128): | |
num_records = len(data) | |
while True: | |
shuffle(data) | |
for offset in range(0, num_records, batch_size): | |
batch_data = data[offset:offset+batch_size] | |
b_inputs_img = [] | |
b_inputs_imu = [] | |
b_labels = [] | |
for seq in batch_data: | |
inputs_img = [] | |
inputs_imu = [] | |
labels = [] | |
for record in seq: | |
#get image data if we don't already have it | |
if record['img_data'] is None: | |
record['img_data'] = np.array(Image.open(record['image_path'])) | |
inputs_img.append(record['img_data']) | |
labels.append(record['target_output']) | |
inputs_imu.append(record['imu_array']) | |
b_inputs_img.append(inputs_img) | |
b_inputs_imu.append(inputs_imu) | |
b_labels.append(labels) | |
X = [np.array(b_inputs_img), np.array(b_inputs_imu)] | |
y = np.array([b_labels]) | |
yield X, y | |
''' | |
With: | |
batchsize - 128 | |
sequence len - 5 | |
image dim - 160, 120, 3 | |
Error: | |
ValueError: Error when checking input: expected img_in to have 4 dimensions, but got array with shape (128, 5, 120, 160, 3) | |
''' |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment