def visualise_with_quiver(model,
input_images='../data/',
class_type='age',
port=5000):
# classes = np.array(['1-15', '16-20', '21-25', '26-30', '31-35', '36-40', '40-45', '46-50', '51-55', '56-100'])
# if class_type != 'age':
# classes = np.array(['female', 'male'])
classes = [
'1-15', '16-20', '21-25', '26-30', '31-35', '36-40', '40-45', '46-50',
'51-55', '56-100'
]
if class_type != 'age':
classes = ['female', 'male']
print("Starting Quiver on port %i" % port)
from quiver_engine import server
server.launch(
model, # a Keras Model
classes, # list of output classes from the model to present (if not specified 1000 ImageNet classes will be used)
10, # number of top predictions to show in the gui (default 5)
# where to store temporary files generated by quiver (e.g. image files of layers)
temp_folder='./tmp',
# a folder where input images are stored
input_folder=input_images,
# the localhost port the dashboard is to be served on
port=port)
def main(model_path, temp_folder_path, input_folder_path, classes, num_predictions):
model = get_model(model_path)
server.launch(
model=model, # a Keras Model
classes=classes,
top=num_predictions, # number of top predictions to show in the gui (default 5)
temp_folder=temp_folder_path,
input_folder=input_folder_path,
port=5000 # the localhost port the dashboard is to be served on
)
def start_server(self):
server.launch(
self.model, # a Keras Model
# where to store temporary files generatedby quiver (e.g. image files of layers)
temp_folder='./tmp',
# a folder where input images are stored
input_folder='../../data',
# the localhost port the dashboard is to be served on
port=5001)
def explore_run(args):
import tempfile
from quiver_engine import server
from . import predict
model, meta = predict.load(args.modeldir)
server.launch(
model,
classes=meta["classes"],
input_folder=args.imagedir,
temp_folder=tempfile.mkdtemp(prefix="mila_quiver_"),
std=[255, 255, 255],
) # This is a bit of a hack to make quiver scale the images
def main():
parser = ArgumentParser()
parser.add_argument("-m", "--model", dest="model")
parser.add_argument("-q", "--quiver", action="store_true", dest="quiver")
args = parser.parse_args()
model_type = ""
if args.model:
model_type = args.model
model = retrieve_option_model(model_type)
model.load_model()
if args.quiver:
server.launch(
model.model,
input_folder="./examples/test_cut",
)
def runserver(model_path, input_images_path):
try:
model = keras.models.load_model(model_path)
except ValueError:
new_model_path = os.path.abspath(model_path) + ".edited"
shutil.copy(model_path, new_model_path)
f = h5py.File(new_model_path, 'r+')
del f['optimizer_weights']
f.close()
model = keras.models.load_model(new_model_path)
server.launch(
model, # a Keras Model
classes=
'regression', # list of output classes from the model to present (if not specified 1000 ImageNet classes will be used)
# a folder where input images are stored
input_folder=input_images_path,
# the localhost port the dashboard is to be served on
port=5000)
# -*- coding: utf-8 -*-
from keras.applications.vgg16 import VGG16
from quiver_engine import server
model = VGG16()
server.launch(model,
input_folder='./sample_images', temp_folder='./tmp', port=8000)
def launch_server(self, imgFolder="GeneratedImages_Bin2"):
names = ['bc', 'bj', 'bn', 'sa', 'bry', 'brb', 'cm', 'cst', 'cso', 'sl',\
'cbp', 'brc', 'cd', 'ds', 'brp', 'sf1', 'sii1', 'siv1', 'sp1', 'sv1',\
'cca', 'cch', 'cgr', 'cme', 'cpe', 'ahy', 'apacc', 'apr', 'apo', 'are']
server.launch(self.model, classes=names, input_folder=imgFolder)
from models.model2_6_9172 import first_model
from quiver_engine import server
input_shape = (64, 64, 3)
model = first_model(input_shape, 6)
model.load_weights(
'save_weights/model2_6/trash-model-weight-ep-176-val_loss-0.33-val_acc-0.92.h5'
)
server.launch(
model,
classes=['glass', 'cardboared', 'metai', 'paper', 'plasic', 'trash'],
input_folder='hhh')
server.launch(model)
import numpy as np
import time
from keras.preprocessing.image import save_img
from keras.applications.vgg16 import VGG16
from keras.models import load_model
from keras import backend as K
from dataset_loader import DatasetLoader
from constants import *
# Dimensions of the generated pictures for each filter.
img_width = 32
img_height = 32
# Load dataset
# Comment/uncomment to select dataset to use
dataset = DatasetLoader()
# CK Extended no resize
dataset.ck_extended_load_from_save()
classes = CK_EXTENDED_EMOTIONS
# Load model
model = load_model(MODELS_PATH + 'model_ck_extended_inception_v3_1.h5')
print ('[+] Model loaded')
print (model.summary())
from quiver_engine.server import launch
launch(model, classes=classes, input_folder='./imgs')
model = None
parser = argparse.ArgumentParser(description='Remote Driving')
parser.add_argument(
'model',
type=str,
help='Path to model h5 file. Model should be on the same path.')
parser.add_argument(
'image_folder',
type=str,
nargs='?',
default='',
help=
'Path to image folder. This is where the images from the run will be saved.'
)
args = parser.parse_args()
#print(args)
# check that model Keras version is same as local Keras version
f = h5py.File(args.model, mode='r')
model_version = f.attrs.get('keras_version')
keras_version = str(keras_version).encode('utf8')
if model_version != keras_version:
print('You are using Keras version ', keras_version,
', but the model was built using ', model_version)
model = load_model(args.model)
server.launch(model, temp_folder='features', input_folder='img', port=5000)
# quiver engine: https://keplr-io.github.io/quiver/ from keras.applications.vgg16 import VGG16 model = VGG16(weights='imagenet') from quiver_engine.server import launch launch(model, input_folder='./img', port=7000) type(model)
tf.python.control_flow_ops = tf
from keras.models import Sequential, model_from_json, load_model
from keras.optimizers import Adam, SGD
from keras.layers.core import Dense, Activation, Flatten, Dropout
from keras.layers.convolutional import Convolution2D
from keras.layers.pooling import MaxPooling2D
from keras.preprocessing.image import *
from sklearn.utils import shuffle
from sklearn.model_selection import train_test_split
import matplotlib.pyplot as plt
import tables
import sys
from quiver_engine import server
# -------------------------------------
# Compile and train the model
# -------------------------------------
model = load_model('model.h5')
#opt = SGD(lr=0.01, decay=1e-6, momentum=0.9, nesterov=True)
model.summary()
# -------------------------------------
# Evaluate the trained model
# -------------------------------------
server.launch(model, input_folder='./quiver_img/IMG')
model.add(
Convolution2D(nb_filters,
kernel_size[0],
kernel_size[1],
border_mode='valid',
activation='relu',
input_shape=input_shape))
model.add(
Convolution2D(nb_filters,
kernel_size[0],
kernel_size[1],
activation='relu'))
model.add(MaxPooling2D(pool_size=pool_size))
model.add(Dropout(0.25))
model.add(Flatten())
model.add(Dense(128, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(nb_classes, activation='softmax'))
model.compile(loss='categorical_crossentropy',
optimizer='adadelta',
metrics=['accuracy'])
model.load_weights('MNIST_weights.h5')
# launching the visualization server
from quiver_engine import server
server.launch(model, temp_folder='./tmp', input_folder='./', port=7777)
# NOTES for running:
# Must run: pip install quiver-engine
# This script runs well from a .ipynb jupyter notebook, just put into a .py file for git sake.
# Model must be trained on .jpg and only .jpg files will load in the quiver visualization.
#
import tensorflow as tf
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
from keras import backend as K
from keras.models import load_model
from keras.models import Model
from keras.layers import Dense, GlobalAveragePooling2D
from keras.utils.vis_utils import plot_model
import keras.backend
from quiver_engine import server
model = load_model("senator_model.h5")
model.summary()
server.launch(model, temp_folder='./tmp', input_folder='./Images/Senators/')
# pip install quiver_engine from quiver_engine import server server.launch(model) # localhost:5000
# Test quiver-time
import keras as ks
from quiver_engine import server
# Load model
model = ks.models.load_model('models/my_bestmodel-win.h5')
# Launch server
server.launch(model, classes=range(7), input_folder='input/')
#!/usr/bin/env python
import os
from quiver_engine import server
from squeezenet import SqueezeNet
classes = [
'Zero', 'One', 'Two', 'Three', 'Four', 'Five', 'Six', 'Seven', 'Eight',
'Nine'
]
input_shape = (67, 67, 3)
weights_file = './mnist_weights.h5'
model = SqueezeNet(len(classes), input_shape=(input_shape))
model.output
model.compile(loss="categorical_crossentropy",
optimizer='adadelta',
metrics=['accuracy'])
if os.path.isfile(weights_file):
print('Loading weights: %s' % weights_file)
model.load_weights(weights_file, by_name=True)
server.launch(model=model,
classes=classes,
temp_folder='/tmp/quiver',
input_folder='/mnt/data/Development/ros/catkin_ws/images')
kernel_size=(3, 3),
padding='same',
input_shape=(3, 100, 100)))
model.add(Activation('relu'))
model.add(Convolution2D(32, (3, 3)))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
model.add(Convolution2D(64, (3, 3), padding='same'))
model.add(Activation('relu'))
model.add(Convolution2D(64, (3, 3)))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
model.add(Flatten())
model.add(Dense(512))
model.add(Activation('relu'))
model.add(Dropout(0.5))
model.add(Dense(5))
model.add(Activation('sigmoid'))
# let's train the model using SGD + momentum (how original).
sgd = SGD(lr=0.001, decay=1e-6, momentum=0.9, nesterov=False)
model.compile(loss='binary_crossentropy', optimizer=sgd, metrics=['accuracy'])
model.load_weights("weights.hdf5")
server.launch(model, input_folder='./', temp_folder='./filters')
def visualize_feature_map(model_name):
model, input_size = model_builder(model_name)
hook_list = register_hook(model)
server.launch(model, hook_list, input_folder="./data/Cat", image_size=input_size, use_gpu=False)
def launch_quiver(model_path):
model = load_model(model_path)
server.launch(model,
port=8015,
input_folder='./imgs',
classes=['pos', 'neg'])
#
# Einsatz von Quiver zur interaktiven Visualisierung eines VGG16 Modells
#
from quiver_engine import server
from keras.applications import VGG16
global model
model = VGG16(weights="imagenet")
model.summary()
# Struktur des Modells
print(model.to_json())
# Quiverboard wird gestartet
server.launch(model,
top=10,
temp_folder="./tmp",
input_folder='./imgs',
port=12345)
model = Sequential()
model.add(Convolution2D(32, kernel_size=(3, 3),padding='same',input_shape=(3 , 100, 100)))
model.add(Activation('relu'))
model.add(Convolution2D(32, (3, 3)))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
model.add(Convolution2D(64,(3, 3), padding='same'))
model.add(Activation('relu'))
model.add(Convolution2D(64, (3, 3)))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
model.add(Flatten())
model.add(Dense(512))
model.add(Activation('relu'))
model.add(Dropout(0.5))
model.add(Dense(5))
model.add(Activation('sigmoid'))
# let's train the model using SGD + momentum (how original).
sgd = SGD(lr=0.001, decay=1e-6, momentum=0.9, nesterov=False)
model.compile(loss='binary_crossentropy', optimizer=sgd, metrics=['accuracy'])
model.load_weights("weights.hdf5")
server.launch(model,input_folder='./',temp_folder='./filters')
# Test quiver-time
import keras as ks
from quiver_engine import server
import sys
# Load model
if sys.version_info.major == 2:
model = ks.models.load_model('models/my_bestmodel.h5')
elif sys.version_info.major == 3:
model = ks.models.load_model('models/my_bestmodel_py3.h5')
# Names of classes of PAMAP2
classes = [
'lying', 'sitting', 'standing', 'walking', 'cycling', 'vacccuum', 'ironing'
]
# Launch server
server.launch(model, classes=classes, top=7, input_folder='input/')
import keras.applications as apps from quiver_engine.server import launch model = apps.vgg16.VGG16() launch(model, input_folder=".")
model_disk = load_model('model/2017-07-09-14-53-loss-decrease-171-0.89.hdf5',
custom_objects={'my_hinge': my_hinge, 'new_smooth': new_smooth})
weights_disk = model_disk.get_weights()
multitask_model.set_weights(weights_disk)
"""
# multitask_model = load_model('model/2017-07-09-14-53-loss-decrease-171-0.89.hdf5',
# custom_objects={'my_hinge': my_hinge, 'new_smooth': new_smooth})
multitask_model = load_model(
'model/2017-07-17-18-43-epoch-16-loss-6.78-saved-all-model.hdf5',
custom_objects={
'my_hinge': my_hinge,
'new_smooth': new_smooth
})
launch(multitask_model, input_folder='./img', port=5000)
# all_imgs = glob.glob('/home/yuquanjie/Documents/shumei_crop_center_test/' + '*.jpg')
all_imgs = glob.glob('/home/yuquanjie/Documents/icdar2017_crop_center/' +
'*.jpg')
# python generator
data_gen_pred = get_pred_img(all_imgs)
while True:
X, img_data = data_gen_pred.next()
# predict
X = np.expand_dims(X, axis=0)
predict_all = multitask_model.predict_on_batch(1 / 255.0 * X)
# 1) classification result
predict_cls = predict_all[0]
# reduce dimension from (1, 80, 80, 1) to (80, 80)
predict_cls = np.sum(predict_cls, axis=-1)
import numpy as np
from quiver_engine import server
from quiver_engine.model_utils import register_hook
from torchvision import models
if __name__ == "__main__":
model = models.resnet18()
hook_list = register_hook(model)
server.launch(model,
hook_list,
input_folder="./data/Cat",
image_size=[250, 250],
use_gpu=False)
weights_path = get_file(
'vgg19_weights_tf_dim_ordering_tf_kernels.h5',
TF_WEIGHTS_PATH,
cache_subdir='models')
else:
weights_path = get_file(
'vgg19_weights_tf_dim_ordering_tf_kernels_notop.h5',
TF_WEIGHTS_PATH_NO_TOP,
cache_subdir='models')
model.load_weights(weights_path)
if K.backend() == 'theano':
convert_all_kernels_in_model(model)
return model
model = VGG19(include_top=True, weights='imagenet')
from quiver_engine import server
server.launch(
model, # a Keras Model
# where to store temporary files generatedby quiver (e.g. image files of layers)
temp_folder='./cache',
# a folder where input images are stored
input_folder='./',
# the localhost port the dashboard is to be served on
port=5000)
filepath=os.path.join(MODEL_DIR, "model-{epoch:02d}.h5"))
model.fit(Xtrain, Ytrain, batch_size=BATCH_SIZE, nb_epoch=NUM_EPOCHS,
validation_split=0.1, callbacks=[checkpoint])
Using TensorBoard and Keras
Keras provides a callback for saving your training and test metrics, as well as activation histograms
for the different layers in your model:
keras.callbacks.TensorBoard(log_dir='./logs', histogram_freq=0,
write_graph=True, write_images=False)
Saved data can then be visualized with TensorBoad launched at the command line:
tensorboard --logdir=/full_path_to_your_logs
Using Quiver and Keras
In Chapter 3 , Deep Learning with ConvNets, we will discuss ConvNets, which are an advanced deep
learning technique for dealing with images. Here we give a preview of Quiver (for more information,
refer to https://github.com/jakebian/quiver ), a tool useful for visualizing ConvNets features in an interactive
way. The installation is pretty simple, and after that Quiver can be used with one single line:
pip install quiver_engine
from quiver_engine import server
server.launch(model)
This will launch the visualization at localhost:5000 .
import os
import argparse
from keras.applications.vgg16 import VGG16
from quiver_engine import server
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument('--img_folder',
type=str,
help='directory with your images')
parser.add_argument('--port', type=int, help='port to serve your quiver')
args = parser.parse_args()
model = VGG16(include_top=True)
tmp_dir = os.path.join(args.img_folder, '.tmp')
if not os.path.exists(tmp_dir): os.makedirs(tmp_dir)
server.launch(model,
input_folder=args.img_folder,
temp_folder=tmp_dir,
port=args.port)
from quiver_engine import server
from tensorflow.keras.models import load_model
from tensorflow.keras.utils import CustomObjectScope
from tensorflow.keras.initializers import glorot_uniform
from keras.preprocessing import image
import numpy as np
model = load_model('C:/Users/jenni/Downloads/artifacts_model.h5')
model.summary()
# temp_folder not working - creates a \tmp file where the project is instead (delete folder if error)
server.launch(model,
classes=['Accepted', 'Crack', 'Chip'],
temp_folder='C:/Data/temp',
input_folder='C:/Data/catdog')
