Day #308 - Display on Rasberry PI

Finally got a 2.8 Inch LCD display for Rasberry PI. Next step is to experiment with the models with real-time situations. This LCD will help to visualize the output. Package installation commands and reference.

	#https://www.waveshare.com/wiki/2.8inch_RPi_LCD_(A)
	pi Display
	============
	git clone https://github.com/waveshare/LCD-show.git
	cd LCD-show/
	/LCD28-show
	To Get back to monitor display
	================================
	cd LCD-show/
	./LCD-hdmi
	Virtual Keyboard
	===================
	#https://raspberrypi.stackexchange.com/questions/41150/virtual-keyboard-activation
	sudo apt-get install matchbox-keyboard
	sudo reboot
	Menu >> Accessories >> Keyboard

view raw RPI_Display.txt hosted with ❤ by GitHub

Happy Learning!!!

Social Media Responsibilities

How do we measure social media impact? What are the pros and cons of Social Media
Pros

• Information Sharing
• Connect with a larger set of population

Cons

• Sharing information without Authenticity
• Motive / Authenticity of people sharing information
• Smiles / Selfies vs Reality of Life
• Manipulative / Biased / Personalized targeted ads
• Exploit human tendencies of feedback / Sensitized to bias 

Social Media Accountability

• Freedom of speech vs Hurting Sentiments
• Consequences of blackmail/threat/bullying through social media contacts
• Consequences of any form of violence instigated through messages
• Validating the facts/claims shared?
• Endorsing political advertisement claims without a moral stand?
• Biased targeting of users?
• Depression / Suicide due to excess usage of social media?
• Freedom of speech vs Authenticity of speech vs Intentions of information shared?

When we can’t even agree on what is real


Keep Thinking!!!

Analyzing top 25 AI companies in 2019

Analyzing top 25 AI companies listed in Link

Happy Learning AI Landscape!!!

Data Analysis vs Forensic Science

Before AI / BI it's about #exploring the data to uncover the #DataInsights. #DataAnalysis is similar to #ForensicScience. Side by Side comparison of both perspectives.

#Data and #Insights sets the #direction for successful #AI / #BI usecases #datascience #bigdata #analytics

Happy Learning!!!

Difference between SQL and NOSQL Systems

Reposting from my two-year-old Quora answer

The Key differences between them lies in the understanding CAP theorem

• Consistency
• Availability
• Partition Tolerance

In layman terms. SQL systems ex-RDBMS will adhere ACID properties (Atomicity, Consistency, Isolation, Durability).

• The datatypes, schema are predefined, You cannot store non-matching datatypes
• To avoid dirty data, systems enforce isolation levels that govern only committed data is read (Consistency)
• Only latest records are available, records at that point in time are not available
• Banking Systems, ordering systems where data needs to consistent will be mostly SQL based systems where consistency is important

No-SQL systems (Not Only SQL)

• The schema is not tightly governed, its flexible you can store different datatypes in same columns
• These may be geographically distributed where data may be synced and eventually be consistent end of day not realtime
• They also support point in time data, data values at a point in time can also be looked up
• Where there is no requirement for consistency we can achieve other 2 Availability and partition tolerance
• Since some of the ACID properties are compromised you will have high availability of this systems

It is more to do with business need to decided SQL or SQL based storage.

Happy Learning!!!

Improving Women Safety

To reduce crime against women more than strengthening laws we need to get to the root cause of issues. We need to analyze the crime data and fix the source of the problem.

We need to analyze the crime patterns based on different aspects to find the underlying patterns.

Pattern vs Solutions

• Correlation with alcohol - How to reduce/limit alcohol consumption 
• Correlation with education - How to reduce dropouts and improve education
• Correlation with income category - Sustainable jobs
• Correlation with marital status - Family aspects
• Correlation to caste - Driven by caste / Unemployment / Dropouts
• Correlation to age group - Social media, porn impact
• Correlation to social behavior - Drugs / partying / Addiction
• Correlation to job type - Government vs Private jobs vs Daily vs Organized Crimes

Education is not limited to a few years. Education is not about degrees. The real purpose of education is to unlearn and relearn things from morality/humanity perspective.

It needs complete society change, not just laws. Let's prepare a safer tomorrow by making the required changes. 

Keep Questioning!!!

Day #307 - Porting Keras to Tensorflow Lite Version

Next Task is to run all the developed models in Pi using Tensorflow Lite. I am using google colab to convert the models into lite version.

	#Step 1
	from google.colab import drive
	drive.mount('/content/drive')
	#Upload data to drive in directory ColabData
	#Step 2
	import tensorflow as tf
	print(tf.__version__)
	#tensorflow.contrib is not there in 2.0
	#!pip install tensorflow==2.0.0
	#Downgrade tensorflow to 1.13.2
	#!pip install tensorflow==1.13.2
	#Step 3
	from tensorflow.contrib import lite
	converter = tf.lite.TFLiteConverter.from_keras_model_file(r'/content/drive/My Drive/ColabData/model_landmark_vgg16.h5')
	tflite_model = converter.convert()
	open(r'/content/drive/My Drive/ColabData/model.tflite','wb').write(tflite_model)

view raw TFLiteConversion.py hosted with ❤ by GitHub

The ported models we will attempt to run in Rasberry PI as next steps

Happy Learning!!!

Day#306 - Express the SQL in pandas, TSQL in Pandas

I wanted to mimic joins, aggregation, sum whatever we do in Database with pandas. A simple storyline of Data Analysis between Employee, Department and Salary using pandas dataframes.

	import pandas as pd
	#Define Data Frames
	Employee = {'name': ['Raj', 'Siva', 'Mike', 'Gopi','New_Joinee'],
	'age': [22,38,26,35,22]}
	dfEmployee = pd.DataFrame(Employee)
	print(dfEmployee)
	#Salary
	Salary = {'name': ['Raj', 'Siva', 'Mike', 'Gopi','Raj', 'Siva'],
	'salary': [2200,3800,2600,3500,7000,5000],
	'Month': ['Jan','Feb','Jan','Jan','Feb','Mar']
	}
	dfSalary = pd.DataFrame(Salary)
	print(dfSalary)
	#Department
	Department = {'name': ['Raj', 'Siva', 'Mike', 'Gopi','NOCODE'],
	'dept': ['IT','AI','HR','DB','NOCODE']}
	dfDepartment = pd.DataFrame(Department)
	print(dfDepartment)
	#Inner Join, Employee and Dept
	print('Outer Join')
	print(pd.merge(dfEmployee, dfDepartment, on='name', how='outer'))
	#Left Join
	print('Left Join')
	print(pd.merge(dfEmployee, dfDepartment, on='name', how='left'))
	#Right Join
	print('Right Join')
	print(pd.merge(dfEmployee, dfDepartment, on='name', how='right'))
	#Inner Join
	print('Inner Join')
	print(pd.merge(dfEmployee, dfDepartment, on='name', how='inner'))
	#Group by
	#Total Salary Group by Employee
	#Do a join
	salarydata = pd.merge(dfEmployee, dfSalary, on='name', how='inner')
	print(salarydata)
	print('Total Paid by Employee')
	#Perform Group By
	print(salarydata.groupby(['name']).sum())
	#Sum
	#Total Salary Paid
	salarydata = pd.merge(dfEmployee, dfSalary, on='name', how='inner')
	Total = salarydata['salary'].sum()
	print('Total Salary Paid')
	print(Total)
	#Min
	#Min Salary paid by employee
	MinimumSalary = salarydata['salary'].min()
	print('MinimumSalary')
	print(MinimumSalary)
	#Max
	#Max Salary paid by employee
	MaxSalary = salarydata['salary'].max()
	print('MaxSalary')
	print(MaxSalary)

view raw Pandas_Joins.py hosted with ❤ by GitHub

Everything can be done in SQL. This is a different approach to it using pandas.

Happy Learning!!!

Day #305 - Loading from Weights file HDF, saved models H5 files

We will look at

• Vanilla Model
• Load preexisting weights HDF5 and Continue
• Load preexisting model H5 and Continue

	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
	from keras import backend as K
	from keras.callbacks import ModelCheckpoint, CSVLogger, EarlyStopping
	import os
	batch_size = 128
	num_classes = 10
	epochs = 5
	log_file_path = r'E:\Landmark\mnist_training_log.log'
	model_save_path = r"E:\Landmark\\mnist.h5"
	weights_filepath="E:\\Landmark\\mnist-weights-improvement-{epoch:02d}.hdf5"
	pre_weights_path = "E:\\Landmark\\mnist-weights-improvement-04.hdf5"
	pre_model_h5_path = "E:\\Landmark\\mnist.h5"
	# input image dimensions
	img_rows, img_cols = 28, 28
	# 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)
	from keras.models import load_model
	def CreateModel():
	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'))
	return model
	def LoadModelfromH5(model_h5_path):
	if os.path.exists(model_h5_path):
	print('Loading Definitions')
	model = load_model(model_h5_path)
	return model
	def LoadModelWeights(pre_weights_path):
	model = CreateModel()
	model.load_weights(pre_weights_path)
	return model
	#Option 1
	#model = CreateModel()
	#Option 2
	#model = LoadModelWeights(pre_weights_path)
	#Option#3
	model = LoadModelfromH5(pre_model_h5_path)
	model.compile(loss=keras.losses.categorical_crossentropy, optimizer=keras.optimizers.Adadelta(),metrics=['accuracy'])
	#Add Early Stop and Checkpoint
	early_stop = EarlyStopping(monitor='val_loss', patience=5, verbose=1)
	checkpoint = ModelCheckpoint(weights_filepath, monitor='val_loss', verbose=1, save_best_only=True, mode='auto')
	csv_logger = CSVLogger(log_file_path, append=False)
	callbacks_list = [checkpoint,early_stop,csv_logger]
	model.fit(x_train, y_train,
	batch_size=batch_size,
	epochs=epochs,
	verbose=1,
	validation_data=(x_test, y_test), callbacks=callbacks_list)
	score = model.evaluate(x_test, y_test, verbose=0)
	print('Test loss:', score[0])
	print('Test accuracy:', score[1])
	model.save(model_save_path)
	import pandas as pd
	import matplotlib.pyplot as plt
	# Plot the Loss
	file_name = log_file_path
	df = pd.DataFrame.from_csv(file_name)
	print(df.head())
	training_loss = df['loss']
	test_loss = df['val_loss']
	print(training_loss)
	print(test_loss)
	epoch_count = range(1, len(training_loss) + 1)
	plt.plot(epoch_count, training_loss, 'r--')
	plt.plot(epoch_count, test_loss, 'b-')
	plt.legend(['Training Loss', 'Test Loss'])
	plt.xlabel('Epoch')
	plt.ylabel('Loss')
	plt.show();

view raw Mnist_load_weights.py hosted with ❤ by GitHub

Results

Option #1 - Vanilla Model

Option #2 - Continue from Saved Weights

Option #3 - Continue from Saved Model H5 File

Happy Learning!!!

Day #304 - Analysis of Deep Fashion Dataset - LandmarkDetection

Three different poses

8 localization points only 4 is not null in all columns
Different visibility value options (0,1,2) - visibility: v=2 visible; v=1 occlusion; v=0 not labeled

One model for each category we need to do
The top-level has 3 generic categories:

• 1: “top” (upper-body clothes such as jackets, sweaters, tees, etc.)
• 2: “bottom” (lower-body clothes such as jeans, shorts, skirts, etc.)
• 3: “long” (full-body clothes such as dresses, coats, robes, etc.)

The implementation is defined in paper - Link
Data - Link

Architecture Implementation

Data Analysis of the Dataset for Non-Zero Columns

• image_name                   0
• landmark_visibility_1        0
• landmark_location_x_1        0
• landmark_location_y_1        0
• landmark_visibility_2        0
• landmark_location_x_2        0
• landmark_location_y_2        0
• landmark_visibility_3        0
• landmark_location_x_3        0
• landmark_location_y_3        0
• landmark_visibility_4        0
• landmark_location_x_4        0
• landmark_location_y_4        0
• landmark_visibility_5    30972
• landmark_location_x_5    30972
• landmark_location_y_5    30972
• landmark_visibility_6    30972
• landmark_location_x_6    30972
• landmark_location_y_6    30972
• landmark_visibility_7    73003
• landmark_location_x_7    73003
• landmark_location_y_7    73003
• landmark_visibility_8    73003
• landmark_location_x_8    73003
• landmark_location_y_8    73003

Non-zero columns

• landmark_visibility_1        0
• landmark_location_x_1        0
• landmark_location_y_1        0
• landmark_visibility_2        0
• landmark_location_x_2        0
• landmark_location_y_2        0
• landmark_visibility_3        0
• landmark_location_x_3        0
• landmark_location_y_3        0
• landmark_visibility_4        0
• landmark_location_x_4        0
• landmark_location_y_4        0

Happy Learning!!!

Day #303 - Model Training Guidelines - Part II

Here we will look at two more additions on top of the previous post 

• Save model h5 file after every run/epoch 
• Add Data batching to run in smaller iterations, Leverage Sequencer

	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
	from keras import backend as K
	from keras.callbacks import ModelCheckpoint, CSVLogger, EarlyStopping
	import os
	import numpy as np
	import math
	batch_size = 128
	num_classes = 10
	epochs = 5
	log_file_path = r'E:\Landmark\mnist_training_log.log'
	model_checkpoint_path = r"E:\Landmark\\mnist.h5"
	model_save_path = r"E:\Landmark\\mnist_model_{}.hd5.h5"
	weights_filepath="E:\\Landmark\\mnist-weights-improvement-{epoch:02d}.hdf5"
	# input image dimensions
	img_rows, img_cols = 28, 28
	# 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)
	#Data Batching
	class Generator(keras.utils.Sequence):
	# Class is a dataset wrapper for better training performance
	def __init__(self, x_set, y_set, batch_size, datacount):
	self.x = x_set
	self.y = y_set
	self.batch_size = batch_size
	self.indices = np.arange(self.x.shape[0])
	self.idx = 0
	self.datacount = datacount
	def __len__(self):
	print('length')
	print(math.ceil(self.datacount/ self.batch_size))
	return math.ceil(self.datacount/ self.batch_size)
	def __getitem__(self, idx):
	print('idx')
	print(idx)
	i1 = idx*self.batch_size
	i2 = (idx+1)*self.batch_size
	print('Start-' + str(i1) + '- End-' + str(i2))
	if(i2 > self.datacount):
	i2 = self.datacount
	batch_x = self.x[i1:i2]
	batch_y = self.y[i1:i2]
	return batch_x, batch_y
	def on_epoch_end(self):
	np.random.shuffle(self.indices)
	#Save Model after every Epoch
	#https://stackoverflow.com/questions/54323960/save-keras-model-at-specific-epochs
	class CustomSaver(keras.callbacks.Callback):
	def on_epoch_end(self, epoch, logs={}):
	#if epoch == 2: # or save after some epoch, each k-th epoch etc.
	self.model.save(model_save_path.format(epoch))
	batch_size = 500
	print('x_train')
	print(len(x_train))
	print('x_test')
	print(len(x_test))
	training_generator = Generator(x_train, y_train, batch_size, len(x_train))
	validation_generator = Generator(x_test, y_test, batch_size, len(x_test))
	from keras.models import load_model
	#Load and Continue Training
	# load weights if it exists
	if os.path.exists(model_checkpoint_path):
	print('Loading Definitions')
	model = load_model(model_checkpoint_path)
	else:
	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'])
	#Add Early Stop and Checkpoint
	early_stop = EarlyStopping(monitor='val_loss', patience=5, verbose=1)
	checkpoint = ModelCheckpoint(weights_filepath, monitor='val_loss', verbose=1, save_best_only=True, mode='auto')
	csv_logger = CSVLogger(log_file_path, append=False)
	saver = CustomSaver()
	callbacks_list = [checkpoint,early_stop,csv_logger,saver]
	model.fit_generator(training_generator, validation_data = validation_generator, epochs = 10, callbacks=callbacks_list)
	score = model.evaluate(x_test, y_test, verbose=0)
	print('Test loss:', score[0])
	print('Test accuracy:', score[1])
	model.save(model_save_path)
	import pandas as pd
	import matplotlib.pyplot as plt
	# Plot the Loss
	file_name = log_file_path
	df = pd.DataFrame.from_csv(file_name)
	print(df.head())
	training_loss = df['loss']
	test_loss = df['val_loss']
	print(training_loss)
	print(test_loss)
	epoch_count = range(1, len(training_loss) + 1)
	plt.plot(epoch_count, training_loss, 'r--')
	plt.plot(epoch_count, test_loss, 'b-')
	plt.legend(['Training Loss', 'Test Loss'])
	plt.xlabel('Epoch')
	plt.ylabel('Loss')
	plt.show();

view raw Mnist_Guidelines.py hosted with ❤ by GitHub

This is a template code. This can be customized for larger datasets

Happy Learning!!!

Project Learning Notes

Tracking, Counting has always been quite interesting topic for sometime. Explored this codebase link

	#https://github.com/poojavinod100/People-Counting-Crowd-Density-Detection/blob/master/people_counter.py
	#pip install imutils
	#pip install CMake
	#pip install dlib
	python people_counter.py --prototxt mobilenet_ssd/MobileNetSSD_deploy.prototxt --model mobilenet_ssd/MobileNetSSD_deploy.caffemodel --output output/webcam_output.avi

view raw commands.txt hosted with ❤ by GitHub

I liked the approach of directionality based tracking. This is very needed for directionality based counting. Hoping to reuse / implement it in people counting scenarios.

My perspective is

• Tracking by Sampling Frames (Reduce Load)
• Use Euclidean and other attributes to track/match
• Evaluate existing tracking built in OpenCV (Again these need frame by frame tracking)

Happy Learning!!!

Day #302 - Keras Best Practices during Training

In this post we take the raw version of code and add below features in code

• Adding Checkpoint
• Adding Logging
• Plot Results
• Restart Training from Checkpoint
• Early Stopping

	#Base code - https://keras.io/examples/mnist_cnn/
	#Added Features
	#========================
	# Adding Checkpoint
	# Adding Logging
	# Plot Results
	# Restart Training from Checkpoint
	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
	from keras import backend as K
	from keras.callbacks import ModelCheckpoint, CSVLogger, EarlyStopping
	import os
	batch_size = 128
	num_classes = 10
	epochs = 5
	log_file_path = r'E:\Landmark\mnist_training_log.log'
	model_checkpoint_path = r"E:\Landmark\\mnist.h5"
	model_save_path = r"E:\Landmark\\mnist.h5"
	weights_filepath="E:\\Landmark\\mnist-weights-improvement-{epoch:02d}.hdf5"
	# input image dimensions
	img_rows, img_cols = 28, 28
	# 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)
	from keras.models import load_model
	#Load and Continue Training
	# load weights if it exists
	if os.path.exists(model_checkpoint_path):
	print('Loading Definitions')
	model = load_model(model_checkpoint_path)
	else:
	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'])
	#Add Early Stop and Checkpoint
	early_stop = EarlyStopping(monitor='val_loss', patience=5, verbose=1)
	checkpoint = ModelCheckpoint(weights_filepath, monitor='val_loss', verbose=1, save_best_only=True, mode='auto')
	csv_logger = CSVLogger(log_file_path, append=False)
	callbacks_list = [checkpoint,early_stop,csv_logger]
	model.fit(x_train, y_train,
	batch_size=batch_size,
	epochs=epochs,
	verbose=1,
	validation_data=(x_test, y_test), callbacks=callbacks_list)
	score = model.evaluate(x_test, y_test, verbose=0)
	print('Test loss:', score[0])
	print('Test accuracy:', score[1])
	model.save(model_save_path)
	import pandas as pd
	import matplotlib.pyplot as plt
	# Plot the Loss
	file_name = log_file_path
	df = pd.DataFrame.from_csv(file_name)
	print(df.head())
	training_loss = df['loss']
	test_loss = df['val_loss']
	print(training_loss)
	print(test_loss)
	epoch_count = range(1, len(training_loss) + 1)
	plt.plot(epoch_count, training_loss, 'r--')
	plt.plot(epoch_count, test_loss, 'b-')
	plt.legend(['Training Loss', 'Test Loss'])
	plt.xlabel('Epoch')
	plt.ylabel('Loss')
	plt.show();

view raw Keras_Best_Practices.py hosted with ❤ by GitHub

Run #1 Output (10 Epochs)

Run #2 Continue with Existing Weights (5 Epochs)

Happy Learning!!!

Day #301 - Data Batching in Keras

This post is about custom data batching using Keras. Here we override the methods of inbuilt sequence. The below example is with dummy data generation, data splitting and fetching the batch of records.

	#Generate dummy data
	import pandas as pd
	import numpy as np
	#Generate 250 Records and Split into 6 Columns
	df = pd.DataFrame(np.random.randint(0,100,size=(250, 6)), columns=list(['X1','X2','X3','X4','Y1','Y2']))
	print(df.head())
	print(df.count())
	#Split into X and Y
	X = df[['X1','X2','X3','X4']]
	Y = df[['Y1','Y2']]
	print(X.head())
	print(Y.head())
	#Split data into train and test
	from sklearn.cross_validation import train_test_split
	x_train, x_test, y_train, y_test = train_test_split(X, Y, test_size=0.2, random_state=2)
	import keras
	import math
	class Generator(keras.utils.Sequence):
	# Class is a dataset wrapper for better training performance
	def __init__(self, x_set, y_set, batch_size, datacount):
	self.x = x_set
	self.y = y_set
	self.batch_size = batch_size
	self.indices = np.arange(self.x.shape[0])
	self.idx = 0
	self.datacount = datacount
	def __len__(self):
	print('length')
	print(math.ceil(self.datacount/ self.batch_size))
	return math.ceil(self.datacount/ self.batch_size)
	def __getitem__(self, idx):
	print('idx')
	print(idx)
	i1 = idx*self.batch_size
	i2 = (idx+1)*self.batch_size
	print('Start-' + str(i1) + '- End-' + str(i2))
	if(i2 > self.datacount):
	i2 = self.datacount
	batch_x = self.x[i1:i2]
	batch_y = self.y[i1:i2]
	return batch_x, batch_y
	def on_epoch_end(self):
	np.random.shuffle(self.indices)
	batch_size = 10
	print('x_train')
	print(x_train['X1'].count())
	print('x_test')
	print(x_test['X1'].count())
	training_generator = Generator(x_train, y_train, batch_size, x_train['X1'].count())
	validation_generator = Generator(x_test, y_test, batch_size, x_test['X1'].count())
	print('training_generator')
	#Compute batches for training data
	num_batches_train = x_train['X1'].count()/batch_size
	for batch_id in range(int(num_batches_train)):
	print('batch_id')
	print(batch_id)
	print(training_generator.__getitem__(batch_id))
	print('validation_generator')
	#Compute batches for test data
	num_batches_test = x_test['X1'].count()/batch_size
	for batch_id in range(int(num_batches_test)):
	print('batch_id')
	print(batch_id)
	print(validation_generator.__getitem__(batch_id))
	#Model Architecture, Layers, Compile
	#Model fitgenerator
	#Model Save Checkpoint

view raw Data_Batching.py hosted with ❤ by GitHub

Other strategies

• Databases -> CSV 50K Data Chunks Records -> Training and Save Checkpoint
• Checkpoint to Save for Each run and reuse for next 50K Chunk of Data

This is a classic data fetching solution. Database can store millions of records. We can fetch each batch, export it to a CSV and use each chunk, train and save checkpoint and continue further for next run.

Happy Learning!!!

Day #300 - Lessons Learnt from Multi-Label Classification

Today is 300th Post on Data Science. It has been a long journey. Still I feel there is a lot more to catchup. Keep Learning, Keep Going.

There are different tasks involved

1. Data Collection - Fatkun Batch Download Image chrome extension to download images
2. Script to reshape images and store in a standard format
3. Simple DB script to update and prepare data

	import pyodbc
	import os
	files = os.listdir(r'E:\Multi_Label\Input_Data\T-Shirt_Jeans')
	base_filepath = r'E:\Multi_Label\Input_Data\T-Shirt_Jeans'
	cnxn = pyodbc.connect(r'DRIVER={SQL SERVER};SERVER=XXXXX\SQLEXPRESS;DATABASE=DataGeneration;Trusted_Connection=yes;')
	cursor = cnxn.cursor()
	for file in files:
	filepath = base_filepath + '\\' + file
	cursor.execute("insert into Dataset([FileName],[Cap],[Jeans],[Jacket],[TShirt],[Shirt],[Pants]) values (?,0,1,0,1,0,0)",filepath)
	cnxn.commit()
	cnxn.close()

view raw pyodbc_insert.sql hosted with ❤ by GitHub

4. This base implementation was useful for model implementation link
5. Data Test Results

	#Test Code
	import cv2
	import os
	import numpy as np
	from keras.models import save_model, load_model
	test_dataset = r'E:\Multi_Label\Test'
	model = load_model(r'E:\Multi_Label\\model_multi_label.h5')
	print(model.summary())
	test_images = []
	arr = os.listdir(test_dataset)
	files = []
	for file in arr:
	path = test_dataset + '\\'+file
	img = cv2.imread(path,1)
	img = cv2.resize(img,(256,256))
	test_images.append([np.array(img)])
	files.append(path)
	i = 0
	for data in test_images:
	test_img_data = np.array(data).reshape(-1,256,256,3)
	result = model.predict(test_img_data)
	#print(result)
	item = ['1-Cap','2 Jeans','3 - Jacket','4-TShirt','5-Shirt','6-Pants']
	values = []
	values.append(result[0][0])
	values.append(result[0][1])
	values.append(result[0][2])
	values.append(result[0][3])
	values.append(result[0][4])
	values.append(result[0][5])
	print(files[i])
	if(float(result[0][0])>0.5):
	print('Cap')
	if(float(result[0][1])>0.5):
	print('Jeans')
	if(float(result[0][2])>0.5):
	print('Jacket')
	if(float(result[0][3])>0.5):
	print('Tshirt')
	if(float(result[0][4])>0.5):
	print('Shirt')
	if(float(result[0][5])>0.5):
	print('Pants')
	#print(values)
	i = i+1
	#E:\Multi_Label\Test\Cap_Jean.jpeg
	#Cap
	#Jeans
	#E:\Multi_Label\Test\Cap_Jean_2.jpeg
	#Cap
	#Jeans
	#E:\Multi_Label\Test\Cap_Jean_3.jpeg
	#Cap
	#Jeans
	#E:\Multi_Label\Test\Jacket_5.jpg
	#Cap
	#Jeans
	#E:\Multi_Label\Test\Jacket_WM2.jpg
	#Cap
	#Jeans
	#Jacket
	#E:\Multi_Label\Test\Jacket_WM3.jpg
	#Cap
	#Jeans
	#Jacket
	#E:\Multi_Label\Test\Jacket_WM4.jpg
	#Cap
	#Jeans
	#E:\Multi_Label\Test\Jacket_WML1.jpg
	#Cap
	#Jeans
	#E:\Multi_Label\Test\T_Shirt_Target.jpg
	#Cap
	#Jeans
	#E:\Multi_Label\Test\T_Shirt_Target1.jpg
	#Cap
	#Jeans

view raw Test_Results.py hosted with ❤ by GitHub

Happy Learning!!!

Amazon Go /Big Basket Smart Machine - Tech Analysis

Amazon Go

• QR Code for user account linking
• Pick items, Auto detected
• Multiple validation points (Video Object Detection, Shelf Weight sensor-based confirmation, RFID reads etc)
• Multiple RFID readers to capture item movement across ISLEs

Cons

• Massive surveillance 
• Real-time computation

IMO, Big Basket Smart Machine is also similar to the implementation

• QR Code for user account linking
• Pick and Buy
• Mobile App integration 
• Weight Sensors used to detect Shelf Item Quantity
• Unique Items in each row. It is not mixed

Since its a standalone machine, it will not need any tracking with RFID / Video camera. Only Weight sensors sufficient and it is a single person operated at a time.

Happy Learning!!!

Day #299 - OpenVino Python Code for Faces - Pedestrain - Attributes

Hope this helps for other developers trying out executing intel openvino models in python

	#https://github.com/wangxiao5791509/Pedestrian-Attribute-Recognition-Paper-List
	import cv2
	import numpy as np
	import sys
	import time
	import os
	FACE_DATA_DIR = '/home/upsquared/Documents/projects/Code/faces'
	DATA_DIR = '/home/upsquared/Documents/projects/Code/samples'
	RESULTS_DIR = '/home/upsquared/Documents/projects/Code/results'
	ATTRIBUTES_RESULTS_DIR = '/home/upsquared/Documents/projects/Code/attribute_results'
	def Detect_Faces():
	attr_bin = '/opt/intel/computer_vision_sdk/deployment_tools/intel_models/face-detection-retail-0004/FP32/face-detection-retail-0004.bin'
	attr_xml = '/opt/intel/computer_vision_sdk/deployment_tools/intel_models/face-detection-retail-0004/FP32/face-detection-retail-0004.xml'
	ped_net = cv2.dnn.readNet(attr_bin, attr_xml)
	ped_net.setPreferableTarget(cv2.dnn.DNN_TARGET_CPU)
	print("Models loaded")
	files = os.listdir(FACE_DATA_DIR)
	#get all files in directory
	#loop through for all files
	i = 0
	for file in files:
	filepath = FACE_DATA_DIR + '//'+ file
	print(filepath)
	rawframe = cv2.imread(filepath)
	frame = cv2.resize(rawframe, (300,300))
	#https://docs.openvinotoolkit.org/latest/_models_intel_person_detection_retail_0013_description_person_detection_retail_0013.html
	try:
	blob = cv2.dnn.blobFromImage(frame,size=(300,300),ddepth=cv2.CV_8U)
	ped_net.setInput(blob)
	out = ped_net.forward()
	predictions = []
	for detection in out.reshape(-1,7):
	image_id,label,conf,x_min,y_min,x_max,y_max = detection
	print(conf)
	#print(label)
	if conf > 0.5:
	predictions.append(detection)
	#print(predictions)
	print(len(predictions))
	for detection in predictions:
	confidence = float(detection[2])
	xmin = int(detection[3]*frame.shape[1])
	ymin = int(detection[4]*frame.shape[0])
	xmax = int(detection[5]*frame.shape[1])
	ymax = int(detection[6]*frame.shape[0])
	print(xmin,ymin,xmax,ymax)
	cv2.rectangle(frame,(xmin,ymin),(xmax,ymax),color=(0,255,0))
	#Crop and Save
	cv2.imshow("Result",frame)
	cv2.waitKey(0)
	cv2.destroyAllWindows()
	result_filepath = RESULTS_DIR + '//'+ str(i) + '.jpg'
	#write the output in directory
	cv2.imwrite(result_filepath,frame)
	i = i+1
	except:
	print('Error')
	print(frame)
	pass
	def Detect_Pedestrians_Adas():
	attr_bin = '/opt/intel/computer_vision_sdk/deployment_tools/intel_models/pedestrian-detection-adas-0002/FP32/pedestrian-detection-adas-0002.bin'
	attr_xml = '/opt/intel/computer_vision_sdk/deployment_tools/intel_models/pedestrian-detection-adas-0002/FP32/pedestrian-detection-adas-0002.xml'
	ped_net = cv2.dnn.readNet(attr_bin, attr_xml)
	ped_net.setPreferableTarget(cv2.dnn.DNN_TARGET_CPU)
	print("Models loaded")
	files = os.listdir(DATA_DIR)
	#get all files in directory
	#loop through for all files
	i = 0
	for file in files:
	filepath = DATA_DIR + '//'+ file
	print(filepath)
	frame = cv2.imread(filepath)
	#https://docs.openvinotoolkit.org/latest/_models_intel_pedestrian_detection_adas_0002_description_pedestrian_detection_adas_0002.html
	blob = cv2.dnn.blobFromImage(frame,size=(672,384),ddepth=cv2.CV_8U)
	ped_net.setInput(blob)
	out = ped_net.forward()
	predictions = []
	for detection in out.reshape(-1,7):
	image_id,label,conf,x_min,y_min,x_max,y_max = detection
	if conf > 0.5:
	predictions.append(detection)
	#print(predictions)
	print(len(predictions))
	for detection in predictions:
	confidence = float(detection[2])
	xmin = int(detection[3]*frame.shape[1])
	ymin = int(detection[4]*frame.shape[0])
	xmax = int(detection[5]*frame.shape[1])
	ymax = int(detection[6]*frame.shape[0])
	print(xmin,ymin,xmax,ymax)
	pedestrian = frame[ymin:ymax,xmin:xmax]
	result_filepath = RESULTS_DIR + '//'+ str(i) + '.jpg'
	#write the output in directory
	cv2.imwrite(result_filepath,pedestrian)
	cv2.rectangle(frame,(xmin,ymin),(xmax,ymax),color=(0,255,0))
	i = i+1
	cv2.imshow("Result",frame)
	cv2.waitKey(0)
	cv2.destroyAllWindows()
	def Detect_Pedestrians_Retail():
	attr_bin = '/opt/intel/computer_vision_sdk/deployment_tools/intel_models/person-detection-retail-0013/FP32/person-detection-retail-0013.bin'
	attr_xml = '/opt/intel/computer_vision_sdk/deployment_tools/intel_models/person-detection-retail-0013/FP32/person-detection-retail-0013.xml'
	ped_net = cv2.dnn.readNet(attr_bin, attr_xml)
	ped_net.setPreferableTarget(cv2.dnn.DNN_TARGET_CPU)
	print("Models loaded")
	files = os.listdir(DATA_DIR)
	#get all files in directory
	#loop through for all files
	i = 0
	for file in files:
	filepath = DATA_DIR + '//'+ file
	print(filepath)
	frame = cv2.imread(filepath)
	#https://docs.openvinotoolkit.org/latest/_models_intel_pedestrian_detection_adas_0002_description_pedestrian_detection_adas_0002.html
	#blob = cv2.dnn.blobFromImage(frame,size=(384,672),ddepth=cv2.CV_8U)
	#https://docs.openvinotoolkit.org/latest/_models_intel_person_detection_retail_0013_description_person_detection_retail_0013.html
	blob = cv2.dnn.blobFromImage(frame,size=(544,320),ddepth=cv2.CV_8U)
	ped_net.setInput(blob)
	out = ped_net.forward()
	predictions = []
	for detection in out.reshape(-1,7):
	image_id,label,conf,x_min,y_min,x_max,y_max = detection
	#print(conf)
	#print(label)
	if conf > 0.5:
	predictions.append(detection)
	#print(predictions)
	print(len(predictions))
	for detection in predictions:
	confidence = float(detection[2])
	xmin = int(detection[3]*frame.shape[1])
	ymin = int(detection[4]*frame.shape[0])
	xmax = int(detection[5]*frame.shape[1])
	ymax = int(detection[6]*frame.shape[0])
	print(xmin,ymin,xmax,ymax)
	pedestrian = frame[ymin:ymax,xmin:xmax]
	result_filepath = RESULTS_DIR + '//'+ str(i) + '.jpg'
	#write the output in directory
	cv2.imwrite(result_filepath,pedestrian)
	cv2.rectangle(frame,(xmin,ymin),(xmax,ymax),color=(0,255,0))
	i = i+1
	#Crop and Save
	cv2.imshow("Result",frame)
	cv2.waitKey(0)
	cv2.destroyAllWindows()
	def Detect_Attributes():
	attr_bin = '/opt/intel/computer_vision_sdk/deployment_tools/intel_models/person-attributes-recognition-crossroad-0200/FP32/person-attributes-recognition-crossroad-0200.bin'
	attr_xml = '/opt/intel/computer_vision_sdk/deployment_tools/intel_models/person-attributes-recognition-crossroad-0200/FP32/person-attributes-recognition-crossroad-0200.xml'
	ped_net = cv2.dnn.readNet(attr_bin, attr_xml)
	ped_net.setPreferableTarget(cv2.dnn.DNN_TARGET_CPU)
	print("Models loaded")
	files = os.listdir(RESULTS_DIR)
	#get all files in directory
	#loop through for all files
	for file in files:
	try:
	filepath = RESULTS_DIR + '//'+ file
	print(filepath)
	frame = cv2.imread(filepath,1)
	#H = 160
	#W = 80
	h,w = frame.shape[:2]
	if (w>=80 and h>=160):
	#https://docs.openvinotoolkit.org/latest/_models_intel_person_attributes_recognition_crossroad_0230_description_person_attributes_recognition_crossroad_0230.html
	blob = cv2.dnn.blobFromImage(frame,size=(80,160),ddepth=cv2.CV_8U)
	ped_net.setInput(blob)
	print("part 1")
	out = ped_net.forward("453")
	#print(out)
	predictions = []
	for detection in out.reshape(-1,8):
	is_male, has_bag, has_backpack, has_hat, has_longsleeves, has_longpants, has_longhair, has_coat_jacket = detection
	predictions.append(detection)
	print(predictions)
	#print(len(predictions))
	for detection in predictions:
	if(detection[0] > 0.5):
	print('MALE')
	else:
	print('Female')
	if(detection[1] > 0.5):
	print('Has BAG')
	if(detection[2] > 0.5):
	print('has_backpack')
	if(detection[3] > 0.5):
	print('has_hat')
	if(detection[4] > 0.5):
	print('has_longsleeves')
	if(detection[5] > 0.5):
	print('has_longpants')
	if(detection[6] > 0.5):
	print('has_longhair')
	if(detection[7] > 0.5):
	print('has_coat_jacket')
	print("part 2")
	out1 = ped_net.forward("456")
	print(out1)
	predictions = []
	for detection in out1.reshape(-1,2):
	point_with_top_colorx, point_with_top_colory = detection
	predictions.append(detection)
	for detection in predictions:
	print(detection[0])
	print(detection[1])
	print("part 3")
	out2 = ped_net.forward("459")
	print(out2)
	predictions = []
	for detection in out2.reshape(-1,2):
	point_with_bottom_colorx, point_with_bottom_colory = detection
	predictions.append(detection)
	for detection in predictions:
	print(detection[0])
	print(detection[1])
	except:
	pass
	#Detect_Faces()
	#Detect_Pedestrians_Adas()
	#Detect_Pedestrians_Retail()
	Detect_Attributes()

view raw OpenVino_Python_samples.py hosted with ❤ by GitHub

Happy Learning!!!

Day #298 - Data Analysis of PNB Defaulters

Data Source - Link

Data Analysis of PNB Defaulters

Chart #1 - Top 20 States By Company Registration State and Defaulted Amount



Chart #2 - Top 20 States By Defaulters Count





Chart #3 - Top 20 Branches with Maximum Defaulters





Chart #4 - Top 20 Branches with Maximum Defaulters Loan Value





Possible Feature Variables

1. Branch Related Approval Score, Higher Defaulters lower the rating
2. Similar Industry Match Score
3. State Related Scores
4. Connections / Joint ventures in Past with Collapsed Companies
5. Rules for threshold limit based on Industry / State / Branch
6. Multiple Models for ongoing monitoring / performance / social medial trends etc..
7. Build a global model with defaulter list across banks and identify common patterns/modus operandi

Happy Learning!!!

Day #297 - Paper Analysis - WIDER Face and Pedestrian Challenge

WIDER Face and Pedestrian Challenge
Tasks - face detection, pedestrian detection, person search
Dataset - WIDER Pedestrian Track - 20,000 images. From surveillance cameras, driving vehicles

Face Detection

• Approach 1 -  single stage detector with the network structure based on RetinaNet [7] and FAN - Face attention network
• Approach 2 - two-stage face detector following Faster R-CNN [12] and FPN Feature pyramid networks [13] framework
• Approach 3 - a two-stage face detection framework. RetinaNet [7] and RefineDet [15]. The team uses two-stage classification and regression to improve the accuracy of classification

PEDESTRIAN DETECTION TRACK

• Approach 1 - basic detection framework of the champion is Cascade R-CNN. Five models are ensembled: ResNet50 [18], DenseNet-161 [19], 197 SENet-154 [20] and two ResNext-101 [21] models.
• Approach 2 - The second team uses FPN [13] and Faster R-CNN [12] as the basis of their detection framework
• Approach 3 - The team at the third place uses Cascade R-CNN [16] as the detection framework

PERSON SEARCH TRACK

• Approach 1 - The winning team designs a cascaded model that utilizes both face and body features for person search. (1) The face detector used here is MTCNN [26] trained on WIDER FACE [4]. (2) The face recognition model backbones include ResNet [18], InceptionResNet-v2 [27], DenseNet [19], DPN and MobiletNet [28]. (3) The Re-ID backbones include ResNet=50, ResNet-101, DenseNet-161 and DenseNet-201
• Approach 2 - The solution is decomposed into two stages - the first stage is to retrieve faces, and the second stage is to retrieve the bodies. Finally, the retrieval results of the two stages are combined as the ranking result. (1) Face Detection. The face detector used here are PCN [29] and MTCNN [26]. (2) Face Retrieval. A second-order networks [30], [31], [32] (ResNet34 as backbone) trained on VGGFace2 [33] with softmax loss and ring loss [34] is used here
• Approach 3 -  In the first step, the face in the query is used to search persons, whose faces can be detected, by face recognition. Then these images are further used to search again in all candidate images by person reidentification feature to get the final result

Happy Learning!!!

Day #296 - TensorflowLite

TensorflowLite models run in gmail, google photos, google assistant, google photos etc..

Advantages

• Low Latency
• No Data connection required
• On device sensors

Key Points

• On Device ML on many platforms
• Tensorflow model saved in graph format
• Converted to Lite format

TFLite

• Model compression
• Quantization
• Optimized SIMD Kernels
• Converter to generate model. Interpreter to run models

Benefits

• Cross-Platform deployment
• Inference speed increases
• Binary size reduction
• Hardware acceleration roadmap

#Build and save keras model
model = build_your_model()
tf.keras.experimental.export_saved_model(model,saved_model_dir)

#convert keras to tensorflow lite model
converter = tf.lite.TFLiteConverter.from_saved_models(saved_model_dir)
#To experiment new feature
converter.experimental_new_converter = True
tflite_model = converter.convert()

Link1, Link2

Improve Performance of models

• Reduce precision of weights (16 bit instead of 32 bit precision) - Quantization
• Pruning - Remove connections during training
• Op Kernels - ARM
• Delegates - GPU (Run on Specialized hardware)

Tensorflow Lite on Micro-controllers is an impressive move. More than mobiles, cross platforms this is a very impressive step.

Happy Learning!!!

Getting new ideas perspectives

Some meaningful tips for new ideas, solutions, creative thinking

	1. Learn a lot of facts
	2. Have deep knowledge of the background material
	3. Spend time thinking about the problem
	4. A curiosity about fundamental characteristics – what makes things tick (big picture and not details)
	5. Strong drive to want to find out the answers
	6. Pick your work colleagues based on their ability to help simulate good ideas
	7. Narrow the scope of the problem if you are not making progress
	8. Moving to a new situation will allow you to change your behavior since there aren’t preconceived expectations of your behavior.
	9. Similarity to a known problem (experience helps)
	10. Structural analysis (break problem into pieces); solve a microproblem first and then build up – divide & conquer
	11. Ask conceptual questions about everyday things
	12. Simplify and deep dive, Spend a lot of time reading
	13. Work in isolation before participating in a group
	14. You got to be a learning machine to improve your thinking
	15. Drive Decision based on facts, behavior, intuition, apply lessons learned in the past
	https://spinlab.me/2017/11/19/isaac-asimov-asks-how-do-people-get-new-ideas/
	https://spinlab.me/2017/07/29/r-w-hamming-on-creativity/
	https://spinlab.me/2017/07/30/claude-shannons-1952-lecture-on-creative-thinking/
	https://qr.ae/TWgnaR
	Another Interesting Read - Idea Generation - https://blog.samaltman.com/idea-generation
	Good Environment to discuss ideas
	Optimistic people
	Think without the constraints
	Good feel for the future
	"Stay away from people who are world-weary and belittle your ambitions"
	"You want to be able to project yourself 20 years into the future, and then think backwards from there. Trust yourself—20 years is a long time; it’s ok if your ideas about it seem pretty radical."
	"Finally, a good test for an idea is if you can articulate why most people think it’s a bad idea, but you understand what makes it good"
	Technical Debt and Product Success - https://medium.com/@romanpichler/technical-debt-and-product-success-42ec1c5718a7

view raw NewIdeas.txt hosted with ❤ by GitHub

mt: @Fisher85M
CC @MikeQuindazzi @antgrasso

What is involved with the #Lean process? [Infographic]#sixsigma #startups #GrowthHacking #education #UX #brainstorming #innovation #agile #CX #IoT

[via @AXA] @ipfconline1 @Fisher85M pic.twitter.com/5LoMRrRKPz

— C-Suite Tech Point (@CsuiteTechPoint) May 28, 2020

Happy Learning!!!

Day #295 - Age - Emotion - Gender Detection Model

It seems I am aging faster than ever.

Deep Learning Model for Age, Gender, Emotion and Real-time implementation. It seems I am agining faster. If it says I am 40, I have just a decade left to code and transition to something else.

Years progressed so fast seems already aged. I hope to code and do something till my day of death. Keep Going...

Happy Learning!!!

Day #294 - Setting up Home Surveillance System

Finally, after a few months, I was able to set up a Home Surveillance System. Person Detection and Real-time alert.

Installation

Demo

Happy Learning!!!

Day #293 - Date with RASA - Chatbot Learning day :)

Found an interesting workshop on end to end demo with RASA.

Training

Demo

	#https://www.youtube.com/watch?v=xu6D_vLP5vY
	#https://github.com/JustinaPetr/Weatherbot_Tutorial
	Rasa based chatbot
	Step 1 - Pre-Requisites
	========================
	1. Clone Project https://github.com/JustinaPetr/Weatherbot_Tutorial
	2. Install Requirements from FULL Code Directory
	cd E:\Code_Repo\Weatherbot_Tutorial\Full_Code
	pip install -r requirements.txt
	3. Download English Spacy model - To parse and get necessary information
	python -m spacy download en
	4. Install npm with node.js. https://www.npmjs.com/get-npm
	https://nodejs.org/dist/v12.13.0/node-v12.13.0-x64.msi
	5. In New Terminal
	npm i -g rasa-nlu-trainer
	Data Annotation - rasa nlu trainer
	Deconstruct into Intent, Entities
	Intent - What it is about
	Entity - Location, Place, Object in Discussion
	Example Messages, Alongside intents, Entities
	Examples
	Greeting
	GoodBye
	Asking
	Step 2 - Data Annotation
	============================
	Step #1 - File data.json
	{
	"rasa_nlu_data":{
	"common_examples":[
	{
	"text":"Hello"
	"intent":"Greet",
	"entities":[]
	}
	{
	"text":"goodbye"
	"intent":"goodbye",
	"entities":[]
	}
	]
	}
	Step #2 - Launch the trainer in Anaconda console
	1. Goto Location E:\Code_Repo\Weatherbot_Tutorial\Full_Code_Latest>
	2. Run Command rasa-nlu-trainer
	3. Custom adding intent and examples
	4. All additional examples present in git code in updated Data.json file
	Step #3 - Train model
	=======================
	1. Configuration File
	- Provide parameters
	- Pipeline - Feature extractors to fetch messages
	- Model save path
	- Data path for annotated data
	{
	"pipeline":"spacy_sklearn",
	"path":"./models/nlu",
	"data":"./data/data.json"
	}
	config_spacy.json file
	2. nlu_model.py file for script for model training
	#import libraries
	from rasa_nlu.converters import load_data
	#load configuration files
	from rasa_nlu.config import RasaNLUConfig
	#load trainer
	from rasa_nlu.model import Trainer
	def train_nlu(data,config,model_sir):
	training_data = load_data(data)
	trainer = Trainer(RasaNLUConfig(config))
	trainer.train(training_data)
	model_directory = trainer.persist(model_dir,fixed_model_name='weathernlu')
	if __name__=='__main__':
	train_nlu('./data/data.json','config_spacy.json','./models/nlu'
	#Run this to train the model
	#Models created in folder directory
	2. Code to test with additional code in nlu_model.py
	#import libraries
	from rasa_nlu.converters import load_data
	#load configuration files
	from rasa_nlu.config import RasaNLUConfig
	#load trainer
	from rasa_nlu.model import Trainer
	from rasa_nlu.model import Metadata, Interpreter
	def train_nlu(data,config,model_sir):
	training_data = load_data(data)
	trainer = Trainer(RasaNLUConfig(config))
	trainer.train(training_data)
	model_directory = trainer.persist(model_dir,fixed_model_name='weathernlu')
	def run_nlu():
	interpreter = interpreter.load('./models/nlu/default/weathernlu',RasaNLUConfig('config_spacy.json'))
	#load the model
	print(interpreter.parse(u"I am planning my holiday to barcelona, I wounder what is the weather out there"))
	if __name__=='__main__':
	run_nlu()
	3. Changes to run for custom packages (Code will run in these versions)
	pip install rasa_core==0.10.3
	pip install rasa-nlu==0.11.5
	Rerun - nlu_model.py file
	Step #4 - Building the conversation
	======================================
	1. Dialogue management will predict action. Domain file. It is yml file.
	2. Key parts are
	slots - placeholders for context of conversation,
	intents - ,
	entities - ,
	templates - ,
	actions -
	3. All details used for predictions
	slot and entities have same attributes - Observations
	template - text responses for users (multiple answers)
	weather_domain.yml
	slots:
	location:
	type:text
	intents:
	- greet
	- goodbye
	- inform
	entities:
	- location
	templates:
	utter-greet:
	- 'Hello, How can i help?'
	utter-goodbye:
	- 'ttyl'
	utter_ask_location:
	-'In what location?'
	actions:
	- utter_greet
	- utter_goodbye
	- utter_ask_location
	- actions.ActionWeather
	Step #5 - Custom Action creation file
	========================================
	actions.py
	from __future__import absolute_import
	from __future__import division
	from __future__import unicode_literals
	from rasa_core.actions.action import Action
	from rasa_core.events import SlotSet
	class ActionWeather(Action):
	def name(self):
	return 'action_weather'
	def run(self,dispatcher,tracker,domain):
	from apixu.client import ApixuClient
	api_key = ""
	#Authentication
	client = ApixuClient(api_key)
	loc = tracker.get_slot('location')
	current = client.getCurrentWeather(q=loc)
	#parse and extract required details
	country = current['location']['country']
	city = current['location']['name']
	condition = current['current']['condition']['text']
	temperature_c = current['current']['temp_c']
	humidity = current['current']['humidity']
	wind_mph = current['current']['wind_mph']
	response = """It is currently {} in {} at the moment. {} {} and wind {}""".format(condition,city,temperature,humidity,wind_mph)
	dispatcher.utter_message(response)
	#custom action
	return [SlotSet('location',loc)]
	#file updated in actions weather_domain.yml
	Step #6 - Story formation
	==========================
	New file Stories.md markdown file in data folder
	Stories.md
	==========
	#story 01
	* greet
	- utter_greet
	## story 02
	* goodbye
	- utter_goodbye
	##story 03
	* inform
	- utter_ask_location
	##story 04
	* inform
	- action_weather
	Step #7
	========
	Start online session
	using train_init.py and train_online.py
	train_init.py
	- train dialogue management model
	- agent used to train model
	- keras polcies used to train model
	- data file path
	- augmentation factor to add more stories
	- Save model with persist function
	pip install rasa-nlu==0.13.1
	Modified train_init.py
	=======================
	Step #8
	========
	train_online.py
	- import libraries
	- parser to parse extract features
	- load model
	Retrain Model
	python -m rasa_core.train -s data/stories.md -d weather_domain.yml -o models/dialogue --epochs 300
	Step #9
	=======
	Run online Training
	python -m rasa_nlu.train -c nlu_model_config.yml --fixed_model_name current --data ./data/nlu.md --path models/ --project nlu
	Step #10
	=========
	dialogue_management_model.py
	Final code to demo
	Summary
	========
	- Install requirements from FULL Code only
	- Run code nlu_model.py train block
	- Run code train_init.py
	- Run code train_online.py (Actual Conversations with chatbot)
	- Action_Listen (Wait for input)
	- Experimented the greet - ask - response - quit workflow
	- Run demo, dialogue_management_model.py
	Next Reads
	https://towardsdatascience.com/create-chatbot-using-rasa-part-1-67f68e89ddad
	https://medium.com/analytics-vidhya/learn-how-to-build-and-deploy-a-chatbot-in-minutes-using-rasa-5787fe9cce19
	https://forum.rasa.com/t/what-is-the-recommended-setup-for-production-deployemnt/1882

view raw rasabot.txt hosted with ❤ by GitHub

Happy Learning!!!

Day #292 - Gstream on Windows10

Note - Do "complete" instead of "typical" for both cases for it to work

This link was useful to experiment and follow.

Step 1. Download installer1 from link

Step 2. Download installer2 from link

Install both of them in Windows, Goto Folder - F:\gstreamer\1.0\x86_64\bin

Step #3 - Command
gst-launch-1.0.exe -v ksvideosrc device-index=0 ! video/x-raw, format=YUY2, width=320, heigh=240, framerate=30/1, pixel-aspect-ratio=1/1 ! videoconvert ! autovideosink

Stream Output

Next Post Stream from rasberry pi to windows

Happy Learning!!!

Day #291 - Working with chatterbot - Windows 10

	#pip install --ignore-installed PyYAML
	#pip install chatterbot
	#pip install chatterbot-corpus
	from chatterbot import ChatBot
	from chatterbot.trainers import ChatterBotCorpusTrainer
	from chatterbot.trainers import ListTrainer
	welcomebot = ChatBot("Charlie")
	#The current training method takes a list of statements that represent a conversation
	initialconversation = [
	"Hello",
	"Hi there!",
	"How are you doing?",
	"I'm doing great.",
	"That is good to hear",
	"Thank you.",
	"You're welcome."
	]
	welcomebottrainer = ListTrainer(welcomebot)
	welcomebottrainer.train(initialconversation)
	#The program selects the closest matching response by searching for the closest matching known statement that matches the input
	response = welcomebot.get_response("Hi there")
	print(response)
	response = welcomebot.get_response("Hello")
	print(response)
	coursebot = ChatBot("Charlie")
	coursebottrainer = ListTrainer(coursebot)
	querycourses = [
	"Share courses",
	"We offer courses in AI, ML, Devops, Big Data",
	"Great ",
	"Share me AI Details",
	"It has machine learning basics to advanced models",
	"What about big data",
	"hadoop, streaming tools and kafka",
	"What about Devops",
	"Docker, Kubernetics"
	]
	coursebottrainer.train(querycourses)
	response = welcomebot.get_response("Share courses")
	print(response)
	response = welcomebot.get_response("Share me AI Details")
	print(response)
	paymentbot = ChatBot("Charlie")
	paymentbottrainer = ListTrainer(paymentbot)
	feescourses = [
	"Devops Fees",
	"Online 15K, Offline 10K",
	"AI Feeds ",
	"Online 25K, Offline 10K",
	"Big Data Fees",
	"Online 35K, Offline 10K",
	]
	paymentbottrainer.train(feescourses)
	response = paymentbot.get_response("Devops Fees")
	print(response)
	response = paymentbot.get_response("Big Data Fees")
	print(response)

view raw chattberbot.py hosted with ❤ by GitHub

Happy Learning!!!

Day #290 - Yolo from OpenCV DNN - Windows 10

OpenCV has examples to invoke Deep Network using readNet modules

Key Methods

• cv.dnn.readNet - Load the network
• cv.dnn.NMSBoxes - Non Max Suppression
• cv.dnn.blobFromImage - Input for Deep Network

Steps

	Package Install - pip install opencv-python
	Step #1 - git clone https://github.com/opencv/opencv
	Step #2 - Goto Path - opencv/opencv/tree/master/samples/dnn
	Link https://docs.opencv.org/master/da/d9d/tutorial_dnn_yolo.html
	Step #3 - Steps to Run
	example_dnn_object_detection --config=[PATH-TO-DARKNET]/cfg/yolo.cfg --model=[PATH-TO-DARKNET]/yolo.weights --classes=object_detection_classes_pascal_voc.txt --width=416 --height=416 --scale=0.00392 --input=[PATH-TO-IMAGE-OR-VIDEO-FILE] --rgb
	Step #4 - cd E:\Code_Repo\opencv\samples\dnn
	Step #5 - For Image
	python object_detection.py --config="E:\\Code_Repo\\darkflow\\cfg\\yolo.cfg" --model="E:\\Code_Repo\\darkflow\\yolov2.weights" --classes="E:\\Code_Repo\\darkflow\\cfg\\coco.names" --width=416 --height=416 --scale=0.00392 --input="E:\Car_Detection\Data1\\frame214000.jpg" --rgb
	Step #6 - For Video
	python object_detection.py --config="E:\\Code_Repo\\darkflow\\cfg\\yolo.cfg" --model="E:\\Code_Repo\\darkflow\\yolov2.weights" --classes="E:\\Code_Repo\\darkflow\\cfg\\coco.names" --width=416 --height=416 --scale=0.00392 --input="E:\Car_Detection\\ch10_20190304115220.mp4" --rgb
	Another Good Ref - https://github.com/rdeepc/ExploreOpencvDnn

view raw yolo_from_opencv.txt hosted with ❤ by GitHub

Results

Happy Learning!!!

Day #289 - Example code for Class Creation, Data Persistence, Email and Phone number Validation

Example code for Class Creation, Data Persistence, Email and Phone number validators

	#pip install phonenumbers
	#pip install validate_email
	import phonenumbers
	from phonenumbers import carrier
	from phonenumbers.phonenumberutil import number_type
	from validate_email import validate_email
	import pickle
	userdata = r'E:\Code_Repo\messaging\phonename.dictionary'
	useremail = r'E:\Code_Repo\messaging\phoneemail.dictionary'
	phonename = {}
	phoneemail = {}
	class Person:
	def __init__(self,name,phonenumber,emailaddress):
	#Constructor
	self.name = name
	self.phonenumber = phonenumber
	self.emailaddress = emailaddress
	def validate_phonenumber(self):
	#Phone number validation
	status = carrier._is_mobile(number_type(phonenumbers.parse(self.phonenumber)))
	return status
	def validate_email_addr(self):
	#Email validation
	is_valid = validate_email(self.emailaddress)
	return is_valid
	def Save_data():
	guestuser1 =Person('Raj','+91-7406947660','siva@siva.com')
	guestuser2 =Person('Raja','+91-7406947760','siva1@siva.com')
	print(guestuser1.validate_email_addr())
	print(guestuser1.validate_phonenumber())
	#Save data
	phonename[guestuser1.phonenumber] = guestuser1.name
	phoneemail[guestuser1.emailaddress] = guestuser1.emailaddress
	phonename[guestuser2.phonenumber] = guestuser2.name
	phoneemail[guestuser2.emailaddress] = guestuser2.emailaddress
	#Persist data
	with open(userdata, 'wb') as config_dictionary_file:
	pickle.dump(phonename, config_dictionary_file)
	with open(useremail, 'wb') as config_dictionary_file:
	pickle.dump(phoneemail, config_dictionary_file)
	def Read_data():
	#read data
	with open(userdata, 'rb') as config_dictionary_file:
	userdatavalues = pickle.load(config_dictionary_file)
	for key,datavalues in userdatavalues.items():
	print(key)
	print(datavalues)
	with open(useremail, 'rb') as config_dictionary_file:
	useremailvalues = pickle.load(config_dictionary_file)
	for key,datavalues in useremailvalues.items():
	print(key)
	print(datavalues)
	Save_data()
	Read_data()

view raw validators_data_persistence.py hosted with ❤ by GitHub

Happy Learning!!!

Day #288 - Messaging using pyzmq

pip install pyzmq

Example - python usage

Output

Real world use case - imagezmq: Transporting OpenCV images

Happy Learning!!!

Day #287 - Dlib Custom Detector

Learning's

• Aspect ratio needs to be maintained
• Trained for only one type of Object (Vim Dishwasher)
• Trained with just a few images 30 images of train and 5 images of test
• Environment - Windows 10, After all the Setup and Steps, It would take ~45 mins to develop for label, training, and testing

Data Set

• Custom Shelf and Vim dishwash Detection

Image Pre-requisites

• I resized data to 512 x 512 format to be consistent
• Place all images in train and test directory respectively before next steps

Label Training Data

• Img Lab Installation - Refer previous post
• Use Shift key to select rectangle
• Use Alt D to delete region
• Each tool has its own commands

	Step 1 - Specify XML and image path
	=====================================
	E:\Code_Repo\dlib\tools\imglab\build\Release\imglab.exe -c E:\Code_Repo\dlib_obj_count\ShelfData\train\train.xml E:\Code_Repo\dlib_obj_count\ShelfData\train
	Step 2
	=======
	cd E:\Code_Repo\dlib_obj_count
	Step 3
	=======
	imglab.exe E:\Code_Repo\dlib_obj_count\ShelfData\train\train.xml

view raw trainingsteps.txt hosted with ❤ by GitHub

Label Testing Data

	Step 1
	========
	E:\Code_Repo\dlib\tools\imglab\build\Release\imglab.exe -c E:\Code_Repo\dlib_obj_count\ShelfData\test\test.xml E:\Code_Repo\dlib_obj_count\ShelfData\test
	Step 2
	======
	cd E:\Code_Repo\dlib_obj_count
	Step 3
	======
	imglab.exe E:\Code_Repo\dlib_obj_count\ShelfData\test\test.xml

view raw testingsteps.txt hosted with ❤ by GitHub

Training Code

	#Base code - https://gist.github.com/atotto/c1ccbfa44ee70a476816f6389834945e
	#Minor changes for my requirements
	import os
	import sys
	import dlib
	options = dlib.simple_object_detector_training_options()
	options.add_left_right_image_flips = False
	options.C = 5
	options.num_threads = 2
	options.be_verbose = True
	training_xml_path = r'E:\Code_Repo\dlib_obj_count\ShelfData\train\train.xml'
	testing_xml_path = r'E:\Code_Repo\dlib_obj_count\ShelfData\test\test.xml'
	dlib.train_simple_object_detector(training_xml_path, r'E:\Code_Repo\dlib_obj_count\ShelfData\detector.svm', options)
	print("")
	print("Training accuracy")
	print(dlib.test_simple_object_detector(training_xml_path, r'E:\Code_Repo\dlib_obj_count\ShelfData\detector.svm'))
	print("Testing accuracy")
	print(dlib.test_simple_object_detector(testing_xml_path, r'E:\Code_Repo\dlib_obj_count\ShelfData\detector.svm'))

view raw train_dlib.py hosted with ❤ by GitHub

Test Code

	import dlib
	import cv2
	detector = dlib.simple_object_detector(r'E:\Code_Repo\dlib_obj_count\ShelfData\detector.svm')
	filepath = r'E:\Code_Repo\dlib_obj_count\ShelfData\test\25.png'
	img = cv2.imread(filepath,1)
	dets = detector(img)
	for d in dets:
	cv2.rectangle(img, (d.left(), d.top()), (d.right(), d.bottom()), (0, 0, 255), 2)
	# Display the resulting frame
	cv2.imshow("frame",img)
	cv2.waitKey(0)
	# When everything done, release the capture
	cv2.destroyAllWindows()

view raw test_dlib.py hosted with ❤ by GitHub

Result

Make your own Haar Cascade on Windows | Quick & Simple

Making your own Haar Cascade Intro - OpenCV with Python for Image and Video Analysis 

Happy Learning!!!

Day #286 - Working with imglab Annotation Tool

Working with imglab annotation tool.

	Step 1
	======
	git clone https://github.com/davisking/dlib
	Step 2
	=======
	cd to tools/imglab
	Step 3
	=======
	mkdir build
	cd build
	Step 4
	=======
	cmake ..
	Step 5
	======
	cmake --build . --config Release
	Step 6
	======
	cd E:\Code_Repo\dlib\tools\imglab\build
	Step 7
	======
	E:\Code_Repo\dlib\tools\imglab\build\Release\imglab.exe

view raw imglab.txt hosted with ❤ by GitHub

Happy Learning!!!

Learning Moments

Last three-four days, I was breaking my head for a segmentation task. There were a ton of tutorials. Everywhere I pick a code and ended up not working. Finally, I managed to segment it. When we sit and learn alone defintely there will be moments of long failures. Do whatever you do with a bit of curiosity and interest. Learning is an ongoing habit. We are not used to proper learning with focus, attention, curiosity, passion, and experimentation.  

Happy Learning!!!

Day #285 - Experimenting with Unet Segmentation

U-Net

• Symmetric U-Shape - Convolutions + Poolings
• Up-Convolutions - Upsampled layers
• Encoder / Decoder
• Contraction / Expansion
• Skip Connections to learn pixel information

There are a ton of tutorials out there but it takes time to find to what works for us :) in our environment. I was experimenting on u-net based segmentation past few days. I will share my learnings on what worked for me.

Step 1 - The initial image is

I am interested in segmenting the parts (products)

Step 2 - The first step is to resize the image into 256 x 256 dimension

Step 3 - The Next Step is to binarize the image

This is the source image. The target image is

Step 4 - Tool - I used paint 3D and white brush in it to segment the required parts for my need

Step 5 - Follow the steps and create the train and label (source and segmented image)

Step 6 - Train the model, Got the repo and customized it link

Step 7 - The predictions for the test image are

Next Demo

 Happy Learning!!!

Day #284 - OpenCV Error in Windows server 2012

• Turn windows features on or off
• Skip the roles screen and directly go to Feature screen
• Select "Desktop Experience" under "User Interfaces and Infrastructure"

Think answer was useful link

Happy Learning!!!

Day #283 - Clustering to group similar Images

For large retail datasets, before object detection. Clustering becomes essential to focus on each cluster to take it forward. Today's post is clustering images into similar groups

• Generate Feature Data based on VGG / Resnet
• Cluster them using Kmeans
• Result output to their respective cluster directory

	#Base Code - https://medium.com/@franky07724_57962/using-keras-pre-trained-models-for-feature-extraction-in-image-clustering-a142c6cdf5b1
	#Modified for our custom need
	from keras.preprocessing import image
	from keras.applications.vgg16 import VGG16
	from keras.applications.resnet50 import ResNet50
	import numpy as np
	import os
	from sklearn.cluster import KMeans
	from keras.applications.resnet50 import preprocess_input, decode_predictions
	import shutil
	datadir = r'E:\Code_Repo\Images'
	output_dir = r'E:\Code_Repo\results'
	def createFolder(directory):
	try:
	if not os.path.exists(directory):
	os.makedirs(directory)
	except OSError:
	print('Error: Creating directory. ' + directory)
	def VGG_Cluster(numberofclusters):
	feature_list = []
	model = VGG16(weights='imagenet', include_top=False)
	files = os.listdir(datadir)
	for file in files:
	img_path = datadir+ str('\\') + file
	img = image.load_img(img_path, target_size=(224, 224))
	img_data = image.img_to_array(img)
	img_data = np.expand_dims(img_data, axis=0)
	img_data = preprocess_input(img_data)
	feature = model.predict(img_data)
	feature_np = np.array(feature)
	feature_list.append(feature_np.flatten())
	feature_list_np = np.array(feature_list)
	kmeans = KMeans(n_clusters=numberofclusters, random_state=0).fit(feature_list_np)
	labelresult = kmeans.labels_
	print(kmeans.labels_)
	print(kmeans.cluster_centers_)
	print('VGG Results')
	#Create Directory based on number of clusters
	for i in range(numberofclusters):
	directoryname = output_dir + str('\\') + str(i) + str('\\')
	createFolder(directoryname)
	for i in range(len(files)):
	img_path = datadir+ str('\\') + files[i]
	#Copy image according to the directory
	print(files[i])
	shutil.copy(img_path, output_dir + '\\' + str(labelresult[i]) + '\\')
	print(labelresult[i])
	def Resnet_Cluster(numberofclusters):
	feature_list = []
	model = ResNet50(weights='imagenet', include_top=False)
	files = os.listdir(datadir)
	for file in files:
	img_path = datadir+ str('\\') + file
	img = image.load_img(img_path, target_size=(224, 224))
	img_data = image.img_to_array(img)
	img_data = np.expand_dims(img_data, axis=0)
	img_data = preprocess_input(img_data)
	feature = model.predict(img_data)
	feature_np = np.array(feature)
	feature_list.append(feature_np.flatten())
	feature_list_np = np.array(feature_list)
	kmeans = KMeans(n_clusters=numberofclusters, random_state=0).fit(feature_list_np)
	labelresult = kmeans.labels_
	print(kmeans.labels_)
	print(kmeans.cluster_centers_)
	print('Resnet Results')
	#Create Directory based on number of clusters
	for i in range(numberofclusters):
	directoryname = output_dir + str('\\') + str(i) + str('\\')
	createFolder(directoryname)
	for i in range(len(files)):
	#Copy image according to the directory
	img_path = datadir+ str('\\') + files[i]
	print(files[i])
	shutil.copy(img_path, output_dir + '\\' + str(labelresult[i]) + '\\')
	print(labelresult[i])
	#VGG_Cluster(3)
	Resnet_Cluster(3)

view raw Image_Clustering.py hosted with ❤ by GitHub

Input - Mixed Set of Images

Output 
Cluster 1

Cluster 2

Cluster 3

More Reads - Example (in R)

Happy Learning!!!

Day #282 - Retail Product Detection / Retail Object Detection

Paper #1 - Automatic Detection of Out-Of-Shelf Products in the Retail Sector Supply Chain

Rule-based information system

• OOS Contribution factors - Measurement of product availability, Measurement of shelf availability
• Approach - Radio-Frequency Identification based
• Rule - “IF (a product is fast-moving) AND (has low sales volatility) AND (POS sales = 0 for today) THEN the product is OOS
• IF Fastmoving product sales count is zero then there is a problem
• Detection approach = Historical data -> Patterns -> Rules = Apply on Current Data

Paper #2 - Retail Shelf Analytics Through Image Processing and Deep Learning
Analysis

• Tasks - Automatic product checkout using segmentation, Object detection of products on store shelves
• Approach - Shelf Image -> Detector for regions (Class, BBox, Mask) -> Crop Each Region -> Object -> Feature Extractor -> KNN Classifier

Paper #3 - A deep learning pipeline for product recognition on store shelves
Analysis

• Shelf Image -> Region proposals -> Crop -> Reference Images -> Refinement -> Detection

Paper #4 - Planogram Compliance Checking Based on Detection of Recurring Patterns
Analysis

• Shelf Image -> Region Partition -> Recurring Pattern Detection -> Compliance Checking

High Level Recommendations (Apply Combination of techniques)

• Shelf Image -> Region Partition -> Region proposals -> Detect Recurring Pattern, Reference Images for refinement -> Prediction

Project Analysis 
The Shelf Detector System For Retail Stores Using Object Detection

pip install -r requirements.txt
python train_obj_detector.py testNutella1

Code Details - https://github.com/bobquest33/dlib_obj_count/blob/master/nutella.pdf
Tool Used - https://imglab.in/

Interesting Read
Retail Product Recognition on Supermarket Shelves

Paper #5 - Rethinking Object Detection in Retail Stores
Key Notes

• Simultaneously object localization and counting, abbreviated as Locount
• Algorithms to localize groups of objects of interest with the number of instances
• Most of the state-of-the-art object detectors use non-maximal suppression (NMS) to post-process object proposals to produce final detections

New Approach

• Cascaded localization and counting network (CLCNet)
• Localize groups of objects of interest with the numbers of instances

Dataset

• Grozi-120 dataset
• Freiburg Groceries dataset
• GameStop dataset
• Retail-121 dataset
• Sku110k dataset
• TGFS dataset

Cascade R-CNN [1] proposes a multi-stage object detection architecture, which is formed by a sequence of detectors trained with increasing IoU thresholds

Locount Dataset

• 140 common commodities, including 9 big subclasses
• Cascaded localization and counting network (CLCNet
• count-regression strategy for counting
• count-classification strategy for counting
• Locount to localize groups of objects with the instance numbers, which is more practical in retail scenarios

Happy Learning!!!

Day #281 - Yolo based Object Counting and Duplicate Removal

Today's learning is Yolo based object counting, duplicate removal using intersection over union metric.

	import cv2
	import sys
	import os
	#Windows 10 Setup
	from darkflow.net.build import TFNet
	sys.path.append(r"E:\Code_Repo\darkflow\darkflow")
	options = {
	'model': 'E:\\Code_Repo\\darkflow\\cfg\\yolo.cfg',
	'load': 'E:\\Code_Repo\\darkflow\\yolov2.weights',
	'threshold': 0.14,
	'gpu': 1.0
	}
	tfnet = TFNet(options)
	#https://github.com/tejaslodaya/car-detection-yolo/blob/master/app_utils.py
	def iou(box1, box2):
	"""Implement the intersection over union (IoU) between box1 and box2
	Arguments:
	box1 -- first box, list object with coordinates (x1, y1, x2, y2)
	box2 -- second box, list object with coordinates (x1, y1, x2, y2)
	"""
	# Calculate the (y1, x1, y2, x2) coordinates of the intersection of box1 and box2. Calculate its Area.
	xi1 = max(box1[0], box2[0])
	yi1 = max(box1[1], box2[1])
	xi2 = min(box1[2], box2[2])
	yi2 = min(box1[3], box2[3])
	inter_area = (yi2 - yi1) * (xi2 - xi1)
	# Calculate the Union area by using Formula: Union(A,B) = A + B - Inter(A,B)
	box1_area = (box1[2] - box1[0]) * (box1[3] - box1[1])
	box2_area = (box2[2] - box2[0]) * (box2[3] - box2[1])
	union_area = box1_area + box2_area - inter_area
	# compute the IoU
	iou = inter_area / union_area
	return iou
	def Object_Count_Yolo(imagefilepath):
	coordinates = []
	boxes = []
	boxlabels = []
	finalboxes = []
	finalboxlabels = []
	img = cv2.imread(imagefilepath, cv2.IMREAD_COLOR)
	result = tfnet.return_predict(img)
	for data in result:
	coordinates = [data['topleft']['x'], data['topleft']['y'],data['bottomright']['x'], data['bottomright']['y']]
	boxes.append(coordinates)
	boxlabels.append(data['label'])
	for k in range(0,len(boxes)):
	selectflag = 0
	for m in range(k+1,len(boxes)):
	iouvalue = iou(boxes[k],boxes[m])
	if iouvalue > .7:
	selectflag = 1
	if(selectflag==0):
	finalboxes.append(boxes[k])
	finalboxlabels.append(boxlabels[k])
	return len(finalboxes)
	def diffObjectCount(SourceImagepath, DestinationImagePath):
	#Compute and Find Difference in Count
	#Return %% based on Count Difference
	SourceObjectCount = Object_Count_Yolo(SourceImagepath)
	DestinationObjectCount = Object_Count_Yolo(DestinationImagePath)
	print(SourceObjectCount)
	print(DestinationObjectCount)

view raw Yolo_Based_Object_Counting.py hosted with ❤ by GitHub

Happy Learning!!!

Day #280 - Human detection and Tracking

Project #1 - Human Detection and Tracking 

Overview

• Detecting a human and its face in a given video and storing Local Binary Pattern Histogram
• Recognize them in any other videos
• Local Binary Pattern Histogram - type of visual descriptor, clockwise direction check neighbour values

Execution Steps
Clone the project
Step 1 - python create_face_model.py -i data
Step 2 - python main.py -v video

Project #2 - Person-Detection-and-Tracking  (Pending Execution)

Overview

• The person detection in Real-time is done with the help of Single Shot MultiBox Detector
• Single Shot MultiBox Detector
• Tracking - Kalman Filter is fed with the velocity, position and direction of the person which helps it to predict the future location 

Single Shot MultiBox Detector

• The core of SSD is predicting category scores and box offsets for a fixed set of default bounding boxes using small convolutional filters applied to feature maps
• The key difference between training SSD and training a typical detector that uses region proposals, is that ground truth information needs to be assigned to specific outputs in the fixed set of detector outputs

Execution Steps
Clone the project https://github.com/ambakick/Person-Detection-and-Tracking

Execute - camera.py in Spyder

Project #3 - Tracker Types Demo Project (Pending Execution)

Overview

• Track Multiple faces
• Download and Experiment

Run Below Demos
demo - track multiple faces.py
Multiple_Trackers.py
face_eye.py
distance_to_camera.py

Datasets - Link

Object Motion Detection and Tracking for Video Surveillance
Measuring size and distance with OpenCV
Calculate X, Y, Z Real World Coordinates from Image Coordinates using OpenCV

Happy Learning!!!

Learning and Growth

Every job / role / growth is more about understanding the problem / perspective / business and technical dimensions. AI has a lot of tools, languages, architecture, research insights. It is ongoing learning to evaluate all possible solutions, productize with certain boundaries, keep learning to expand other architectures. Be open to learning, titles/roles are a cascade effect of your consistent efforts.

Keep Going!!!

Day #279- Multi-Object Tracking

Project #1 - vehicle-speed-check

Clone Repository - Link

On Anaconda prompt,
cd vehicle-speed-check
pip install -r requirements.txt
python speed_check.py

Comments - Very good project to get started. The logic of speed computation with respect to frames per second, pixel movement can be reused in other use cases. Use of dlib correlation tracker. The tracking logic can be reused in other similar implementation

Project #2 - Simple Example code (ROI Based)

	#pip uninstall opencv-python
	#pip uninstall opencv-contrib-python
	#pip intall opencv-contrib-python.
	#Learnt from https://www.learnopencv.com/multitracker-multiple-object-tracking-using-opencv-c-python/
	#Rewrote based on my requirements
	import cv2
	import sys
	# Read video
	video = cv2.VideoCapture(r"E:\Code_Repo\vehicle-speed-check\cars.mp4")
	#Create two trackers
	trackers = []
	tracker1 = cv2.TrackerKCF_create()
	trackers.append(tracker1)
	tracker2 = cv2.TrackerKCF_create()
	trackers.append(tracker2)
	tracker3 = cv2.TrackerKCF_create()
	trackers.append(tracker3)
	if not video.isOpened():
	print("Could not open video")
	sys.exit()
	# Read first frame.
	ok, frame1 = video.read()
	height , width , layers = frame1.shape
	new_h=int(height/2)
	new_w=int(width/2)
	frame = cv2.resize(frame1, (new_w, new_h))
	if not ok:
	print('Cannot read video file')
	sys.exit()
	bboxdata = []
	#Select two regions to track
	for i in range(3):
	bbox = cv2.selectROI(frame, False)
	bboxdata.append(bbox)
	print(bbox)
	print(bboxdata)
	# Initialize tracker with first frame and bounding box
	#tracker1 update
	ok1 = trackers[0].init(frame, bboxdata[0])
	#tracker2 update
	ok2 = trackers[1].init(frame, bboxdata[1])
	#tracker3 update
	ok3 = trackers[2].init(frame, bboxdata[2])
	flag = 0
	count = 0
	while True:
	# Read a new frame
	ok1, frame1 = video.read()
	height , width , layers = frame1.shape
	new_h=int(height/2)
	new_w=int(width/2)
	frame = cv2.resize(frame1, (new_w, new_h))
	if not ok1:
	break
	i = 0
	for tracker in trackers:
	result, bbox = tracker.update(frame)
	i = i + 1
	if result:
	# Tracking success
	p1 = (int(bbox[0]), int(bbox[1]))
	p2 = (int(bbox[0] + bbox[2]), int(bbox[1] + bbox[3]))
	cv2.rectangle(frame, p1, p2, (255,0,0), 2, 1)
	else:
	print('Tracking failure for Object ', i)
	#remove from the list
	trackers.remove(tracker)
	#for every object that enters append to this
	if flag==0:
	bbox = cv2.selectROI(frame, False)
	tracker = cv2.TrackerKCF_create()
	tracker.init(frame, bbox)
	trackers.append(tracker)
	flag = 1
	cv2.imshow("Tracking", frame)
	# Exit if ESC pressed
	k = cv2.waitKey(1) & 0xff
	if k == 27: break
	cv2.destroyAllWindows()
	print(trackers)

view raw OpenCV_Custom_Tracker.py hosted with ❤ by GitHub

Project #3 - Another interesting project from Adrian blog 

Cloned the project and executed the demo. This code does not work in windows 10 though. Someone has fixed the code. The working code is in link 

python multi_object_tracking_fast.py --prototxt E:\Code_Repo\multiobject-tracking-dlib\mobilenet_ssd\MobileNetSSD_deploy.prototxt  --model E:\Code_Repo\multiobject-tracking-dlib\mobilenet_ssd\MobileNetSSD_deploy.caffemodel --video E:\Code_Repo\multiobject-tracking-dlib\race.mp4 --output E:\Code_Repo\multiobject-tracking-dlib\race_output_fast.avi

More Reads
Speed Detection

Multiple object tracking using dlib library, Python

Face Tracking in Python using Xailient Face Detector and dlib, Link

Object Tracker - Link

Happy Learning!!!

Day #278 - Object Tracking - TensorFlow Object Counting API

I came across this project. Fantastic work!! The Detection part needs to be finetuned for the Indian scenario, the tracking seems fine performing decently. You can spot a few false positives, Trucks on the other side of the lane are not detected, Indian Trucks are not well recognized. This can be handled by a custom detection model. Overall the tracking and counter approach can be reused in multiple scenarios.

Clone the project - object_counting_api

vehicle_counting.py - Executed this for some of my highway videos.

	is_color_recognition_enabled = 1 # set it to 1 for enabling the color prediction for the detected objects
	roi = 600 # roi line position
	deviation = 2 # the constant that represents the object counting area

view raw code_change_vehicle_counting.py hosted with ❤ by GitHub

The objects passing through the line will be counted and incremented. Minor changes to roi

Output of the same



Happy Learning!!!

Analytics in Elections

If Analytics is used to target people, neutralize opinions, create digital impressions. This would lead to the creation of bias. This is not an ethical use of AI.

Analytics is needed to create affordable health care, forecast economy, provide good insights to develop people, economy jobs. AI should be used in the right sense. AI for politics will benefit in the short term but it is a curse in the long term.

When the truth is neutralized with biased facts the consequences of power with the wrong leaders will be a curse for the future generations

Keep Thinking!!!

The curse of Social Media and State of Education - Dark Future

With the growing amount of selfies, pics data shared across quora, facebook, Instagram, tiktok. We need to think of

• Impact of the content on the psychology of kids
• Emotional Maturity / Bias / Impact on Kids / Children
• Loss of productivity / Distraction
• There is considerably more bad content than the good content
• How does these content fuel for better thinking / better focus
• Who is responsible for moderating all the online content? How do I know as a parent the influence on my Kids
• The state of education/mindset hasn't changed with this cheap data plans. Instead, we started losing our own individual thinking
• This is a follow the crowd pattern approach. Developed countries do not have so much of exams but produce better research and innovation. With the state of education, we produce more failures and ship abroad few successful students
• Changing students to merely solving problems within time constraint does not develop creativity or curiosity
• If 90 kids fail in this education system what do they end up? Delivery boys?
• Population can't be controlled by force. It needs education, awareness, and responsibility

Better metrics for a better future generation and better thinking towards

• How do we make people self-sufficient
• How has education improved scientific thinking
• How has education improved morality
• How has education promoted social responsibility
• Are things affordable for the poor and needy
• What have we done to create ideas at grassroots

Creating a better intelligent generation needs more thoughts than just about GDP

Keep Thinking!!!

The Curse of Cheap Data Plans

Many time I wonder cheap data plans are a curse, not a boom. I see more often these days

• More time I personally spend on Youtube
• Forwards of TikTok/ Halo Status videoes
• Rechecking same repetitive news everywhere

I have lost a lot of sleeping hours. Google Youtube recommendation is the most unfair recommendation. Providing extremely similar recommendations. There is no mix of different sources. Sometimes tailored information is not what we need, we need the raw data.

Too much of personalization is a curse. You will lose yourself biased on your perspectives. Sometimes raw information makes more sense than tailored information.

Escape the Web!!!

Day #277 - Tracking Objects - Deep SORT

What is Deep Sort ?
Simple Online and Realtime Tracking with a Deep Association Metric

How it works ?
It performs Kalman filtering in image space and frame-by-frame data association using the Hungarian method with an association metric that measures bounding box overlap

Paper - Link

	Step #1 - Clone it - https://github.com/nwojke/deep_sort
	Step #2 - Download Data from - https://motchallenge.net/data/MOT16/
	Step #3- Download Data from https://drive.google.com/drive/folders/1VVqtL0klSUvLnmBKS89il1EKC3IxUBVK
	Step #4 - Goto Project Directory in Anaconda Prompt
	Step #5 - python deep_sort_app.py --sequence_dir=C:\Data_ML\MOT16\test\MOT16-06 --detection_file=C:\Data_ML\MOT16\Data_Test/MOT16-06.npy --min_confidence=0.3 --nn_budget=100 --display=True

view raw DeepSort.txt hosted with ❤ by GitHub

Happy Learning!!!

Day # 276 - Segmentation, Age-Gender Estimations

This post is based on current learning's exploring segmentation and age/gender detection.

Experiment #1 - Used the existing model and took faces from edouardjanssens.com for both men / women from 1-100 in increments of 5. The actual result and predicted chart summary

Demo Code

	#http://edouardjanssens.com/art/1-to-100-years/men/
	#http://edouardjanssens.com/art/1-to-100-years/women/
	import cv2
	print(cv2.__version__)
	import glob, os
	from pyagender import PyAgender
	agender = PyAgender()
	def Approach1():
	f = open(r'C:\Data_ML\Data_Scrub\Results.csv','w')
	os.chdir(r"C:\Data_ML\Data_Scrub\Boys")
	for file in glob.glob("*.*"):
	faces = agender.detect_genders_ages(cv2.imread(file))
	#print(faces)
	print("_________________________________________________")
	print(file)
	for data in faces:
	for key in data:
	if(key=='gender' or key == 'age'):
	if key=='gender':
	if data[key] <0.5:
	print(key)
	print('Male')
	val1 = 'Male'
	else:
	print(key)
	print('Female')
	val1 = 'Female'
	if key=='age':
	print(key)
	print(data[key])
	val2= data[key]
	print("_________________________________________________")
	data = file + "," + str(val1) + "," + str(val2) + "\n"
	f.write(data)
	f.close()

view raw Test_age_gender.py hosted with ❤ by GitHub

Experiment #2 - Getting Started with Segmentation, Experimented with Repo - https://github.com/divamgupta/image-segmentation-keras

Few tweaks in code

Demo Code

	#https://github.com/divamgupta/image-segmentation-keras
	import keras_segmentation
	model = keras_segmentation.pretrained.pspnet_50_ADE_20K() # load the pretrained model trained on ADE20k dataset
	# load 3 pretrained models
	out = model.predict_segmentation(
	inp=r"E:\Code_Repo\image-segmentation-keras\sample_images\Data1.jpeg",
	out_fname=r"E:\Code_Repo\image-segmentation-keras\sample_images\out1.png"
	)
	model2 = keras_segmentation.pretrained.pspnet_101_cityscapes()
	out = model2.predict_segmentation(
	inp=r"E:\Code_Repo\image-segmentation-keras\sample_images\Data1.jpeg",
	out_fname=r"E:\Code_Repo\image-segmentation-keras\sample_images\out2.png"
	)
	model3 = keras_segmentation.pretrained.pspnet_101_voc12()
	out = model3.predict_segmentation(
	inp=r"E:\Code_Repo\image-segmentation-keras\sample_images\Data1.jpeg",
	out_fname=r"E:\Code_Repo\image-segmentation-keras\sample_images\out3.png"
	)

view raw segdemo.py hosted with ❤ by GitHub

Happy Learning!!!