Hi Aerial Robotics Guru,
1. Requirements for building tensorflow:
- numpy of pip package.
- mock of pip package.
- Java 8 is required for bazel. (Not required for TF execution)
- bazel is required. (Not required for TF execution)
In addition, patches may be applied to the source code.
https://github.com/naisy/JetsonXavier/blob/JetPack4.0_python3.6/JetPack4.0/python3.6/scripts/build_tensorflow.sh
2. Training MNIST data using LeNet model in Keras:
It seemed that there was no problem as far as I tried.
# remove naisy build tensorflow
pip3 uninstall tensorflow
# install official tensorflow
pip3 install --extra-index-url https://developer.download.nvidia.com/compute/redist/jp40 tensorflow-gpu
# install keras-2.2.0
pip3 install --upgrade keras==2.2.0
- Source code (mnist_lenet.py)
# https://github.com/keras-team/keras/blob/master/examples/mnist_cnn.py
'''Trains a simple convnet on the MNIST dataset.
Gets to 99.25% test accuracy after 12 epochs
(there is still a lot of margin for parameter tuning).
16 seconds per epoch on a GRID K520 GPU.
'''
from __future__ import print_function
import keras
from keras.datasets import mnist
from keras.models import Sequential
from keras.layers import Dense, Dropout, Flatten
from keras.layers import Conv2D, MaxPooling2D, AveragePooling2D
from keras import backend as K
batch_size = 128
num_classes = 10
epochs = 12
# input image dimensions
img_rows, img_cols = 28, 28
def LeNet(input_shape, num_classes):
model = Sequential()
model.add(Conv2D(20, kernel_size=5, strides=1, activation='relu', input_shape=input_shape))
model.add(MaxPooling2D(2, strides=2))
model.add(Conv2D(50, kernel_size=5, strides=1, activation='relu'))
model.add(MaxPooling2D(2, strides=2))
model.add(Dropout(0.25))
model.add(Flatten())
model.add(Dense(500, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(num_classes, activation='softmax'))
model.compile(loss=keras.losses.categorical_crossentropy,
optimizer=keras.optimizers.SGD(),
metrics=['accuracy'])
return model
def default_cnn(input_shape, num_classes):
model = Sequential()
model.add(Conv2D(32, kernel_size=(3, 3), activation='relu', input_shape=input_shape))
model.add(Conv2D(64, (3, 3), activation='relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
model.add(Flatten())
model.add(Dense(128, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(num_classes, activation='softmax'))
model.compile(loss=keras.losses.categorical_crossentropy,
optimizer=keras.optimizers.Adadelta(),
metrics=['accuracy'])
return model
# the data, split between train and test sets
(x_train, y_train), (x_test, y_test) = mnist.load_data()
if K.image_data_format() == 'channels_first':
x_train = x_train.reshape(x_train.shape[0], 1, img_rows, img_cols)
x_test = x_test.reshape(x_test.shape[0], 1, img_rows, img_cols)
input_shape = (1, img_rows, img_cols)
else:
x_train = x_train.reshape(x_train.shape[0], img_rows, img_cols, 1)
x_test = x_test.reshape(x_test.shape[0], img_rows, img_cols, 1)
input_shape = (img_rows, img_cols, 1)
x_train = x_train.astype('float32')
x_test = x_test.astype('float32')
x_train /= 255
x_test /= 255
print('x_train shape:', x_train.shape)
print(x_train.shape[0], 'train samples')
print(x_test.shape[0], 'test samples')
# convert class vectors to binary class matrices
y_train = keras.utils.to_categorical(y_train, num_classes)
y_test = keras.utils.to_categorical(y_test, num_classes)
#model = default_cnn(input_shape, num_classes)
model = LeNet(input_shape, num_classes)
model.fit(x_train, y_train,
batch_size=batch_size,
epochs=epochs,
verbose=1,
validation_data=(x_test, y_test))
score = model.evaluate(x_test, y_test, verbose=0)
print('Test loss:', score[0])
print('Test accuracy:', score[1])
python mnist_lenet.py
Using TensorFlow backend.
x_train shape: (60000, 28, 28, 1)
60000 train samples
10000 test samples
Train on 60000 samples, validate on 10000 samples
Epoch 1/12
2018-10-03 05:34:14.234838: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:857] ARM64 does not support NUMA - returning NUMA node zero
2018-10-03 05:34:14.235162: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1405] Found device 0 with properties:
name: Xavier major: 7 minor: 2 memoryClockRate(GHz): 1.5
pciBusID: 0000:00:00.0
totalMemory: 15.46GiB freeMemory: 9.55GiB
2018-10-03 05:34:14.235332: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1484] Adding visible gpu devices: 0
2018-10-03 05:34:15.064031: I tensorflow/core/common_runtime/gpu/gpu_device.cc:965] Device interconnect StreamExecutor with strength 1 edge matrix:
2018-10-03 05:34:15.064223: I tensorflow/core/common_runtime/gpu/gpu_device.cc:971] 0
2018-10-03 05:34:15.064312: I tensorflow/core/common_runtime/gpu/gpu_device.cc:984] 0: N
2018-10-03 05:34:15.064639: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1097] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 9066 MB memory) -> physical GPU (device: 0, name: Xavier, pci bus id: 0000:00:00.0, compute capability: 7.2)
60000/60000 [==============================] - 12s 196us/step - loss: 1.4815 - acc: 0.5106 - val_loss: 0.3588 - val_acc: 0.9084
Epoch 2/12
60000/60000 [==============================] - 6s 92us/step - loss: 0.4331 - acc: 0.8666 - val_loss: 0.2002 - val_acc: 0.9450
Epoch 3/12
60000/60000 [==============================] - 5s 91us/step - loss: 0.2958 - acc: 0.9104 - val_loss: 0.1520 - val_acc: 0.9561
Epoch 4/12
60000/60000 [==============================] - 5s 91us/step - loss: 0.2391 - acc: 0.9277 - val_loss: 0.1248 - val_acc: 0.9622
Epoch 5/12
60000/60000 [==============================] - 5s 90us/step - loss: 0.2048 - acc: 0.9381 - val_loss: 0.1072 - val_acc: 0.9676
Epoch 6/12
60000/60000 [==============================] - 5s 90us/step - loss: 0.1834 - acc: 0.9453 - val_loss: 0.0963 - val_acc: 0.9724
Epoch 7/12
60000/60000 [==============================] - 5s 89us/step - loss: 0.1656 - acc: 0.9501 - val_loss: 0.0864 - val_acc: 0.9737
Epoch 8/12
60000/60000 [==============================] - 5s 89us/step - loss: 0.1541 - acc: 0.9541 - val_loss: 0.0790 - val_acc: 0.9762
Epoch 9/12
60000/60000 [==============================] - 5s 89us/step - loss: 0.1416 - acc: 0.9572 - val_loss: 0.0738 - val_acc: 0.9776
Epoch 10/12
60000/60000 [==============================] - 5s 88us/step - loss: 0.1339 - acc: 0.9593 - val_loss: 0.0683 - val_acc: 0.9786
Epoch 11/12
60000/60000 [==============================] - 5s 88us/step - loss: 0.1255 - acc: 0.9612 - val_loss: 0.0648 - val_acc: 0.9797
Epoch 12/12
60000/60000 [==============================] - 5s 88us/step - loss: 0.1204 - acc: 0.9641 - val_loss: 0.0614 - val_acc: 0.9807
Test loss: 0.06142731437981129
Test accuracy: 0.9807