Day #117 - Yolo Object Detection

This post is using Yolo for object detection

Step 1 - Download Repo - https://github.com/thtrieu/darkflow
Step 2 - Install Commands - https://github.com/markjay4k/YOLO-series/blob/master/part1%20-%20setup%20YOLO.ipynb

Step 3 - Install Cython

Step 4 - Build Downloaded Code

Step 5 - Made Changes to following lines in code

Sample Example Code

Yolo on Colab - Link

	import cv2
	import sys
	sys.path.append("E:\\darkflow-master\\darkflow-master\\")
	from darkflow.net.build import TFNet
	import matplotlib.pyplot as plt
	options = {
	'model': 'E:\\RNotes\\darkflow-master\\cfg\\yolo.cfg',
	'load': 'E:\\RNotes\\darkflow-master\\bin\\yolov2.weights',
	'threshold': 0.14,
	'gpu': 1.0
	}
	tfnet = TFNet(options)
	img = cv2.imread('E:\\yolo\\Test.jpeg', cv2.IMREAD_COLOR)
	#img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
	# use YOLO to predict the image
	result = tfnet.return_predict(img)
	print(result)
	print(img.shape)

view raw yolo.py hosted with ❤ by GitHub

Happy Learning!!!

Day #116 - Fake news detector

There are different kinds of Fake news

• Manipulating Event reported date, place and parties involved - - Fabricated news everything Fake
• Spreading Hatred against Specific groups
• Hate Messages

Solution Approach

• Step #1 - Using Naive Bayes to classify them Hate messages 
• Step #2 - Using Multiple sources to validate news details (Location, Date, news type verification)
• Step #3 - Validate the parties involved by comparing articles on same news
• Step #4 - On Major variations from reported news alert
• Step #5 - Facebook or google they will have large volumes of corpus to train, find more specific details on Fake news - Origination, Place, Time, ethnicity, Age group, Target Audience, Depending on them you can pre-screen news from such audience

Happy AI Learning!!!

Day #115 - Image Template Comparison

Template Matching, OpevCV3, Python 3 Environment

	#pip install opencv-python==3.3.0.10 opencv-contrib-python==3.3.0.10
	#https://stackoverflow.com/questions/37039224/attributeerror-module-object-has-no-attribute-xfeatures2d-python-opencv-2
	#Updated April 25/2020
	import numpy as np
	import cv2
	import matplotlib.pyplot as plt
	img1 = cv2.imread('D:\\Template\\p23.jpg',0) # queryImage
	img2 = cv2.imread('D:\\TestImage\T3.jpeg',0) # trainImage
	sift = cv2.xfeatures2d.SIFT_create()
	# find the keypoints and descriptors with SIFT
	kp1, des1 = sift.detectAndCompute(img1,None)
	kp2, des2 = sift.detectAndCompute(img2,None)
	# BFMatcher with default params
	bf = cv2.BFMatcher()
	matches = bf.knnMatch(des1,des2, k=2)
	# Apply ratio test
	good = []
	for m,n in matches:
	if m.distance < 0.7*n.distance:
	good.append([m])
	img3 = cv2.drawMatchesKnn(img1,kp1,img2,kp2,good,None,flags=2)
	plt.imshow(img3),plt.show()
	print('Number of matches - SIFT')
	print(len(good))

view raw comparison.py hosted with ❤ by GitHub

Happy Learning!!!

Day #114 - OCR with Tesseract

Architecture for Tesseract link

Step 1 - Download Latest version for windows link

Step 2 - Pdf to Image Conversion using ImageMagik (Refer Previous Posts)

convert.exe -density 300 -trim D:\OCR\TestData\DEC17.pdf -quality 100 D:\OCR\TestData\DEC17.jpg

convert.exe -density 300 -trim D:\OCR\TestData\DEC17.pdf -quality 60 D:\OCR\TestData\DEC17.jpg

Step 3 - Tesseract Options using LSTM Networks

#use LSTM Mode
tesseract.exe --oem 2 D:\OCR\TestData\DEC17-0.jpg D:\OCR\TestData\DEC171.csv

tesseract.exe --oem 2 D:\OCR\TestData\DEC17-0.jpg D:\OCR\TestData\ tsv

Intelligent Photo OCR that reads better than you (Or not)
Deep Learning Fun - Crazy Food Image Classifier
OCR and OpenCV with RNN

Ready-to-use OCR with 40+ languages supported including Chinese, Japanese, Korean and Thai.
Build your own OCR(Optical Character Recognition) for free
An OCR System for recognition of Urdu text in Nastaliq Font
Optical Character Recognition

• Image Thresholding
• Contour formation
• Detecting individual characters and forming a Region of Image (ROI) around detected characters.
• Passing the ROIs through our trained model and finding out the predicted character
• Saving/Appending this character in a string, which is basically going to be your output.

Happy Learning!!!

Reading Research papers

Very insightful, practical and detailed. Its all about focus, repetitive efforts and passion to learn. Amazing lecture.

From Siraj Session - Link

Goal Oriented Reading Strategy

Phase I

• Read title, Abstract 
• Use as Overview
• Skim through sections / sub sections
• No math in Pass I
• Correlate to known learning's
• Related Papers

Phase II

• Understand Mathematics
• Get Concept of Maths Formula
• Evaluate reports, repeatable results
• Download code repository
• Replicate results
• Additional resources on web to summarize texts
• Output - Notes, Helper Images

Phase III

• Maths
• Every detail of Math
• Break down equations
• Wikipedia references
• Replicate paper programmatically using equations / settings

Key is 'Never Give Up', 'Turn your frustrations into Fuel', 'Ask for Help'

How to Write a Research Paper

From Siraj Session - Link

• Remind to Stay Positive and Belief
• Start with Questions to arrive at Topic
• Broad / Specific
• Find answers for those questions
• Well articulated and laser focussed on solution for problem
• Collect data on topic
• Remember to search through variey of sources, critical in assessment
• Start Learning from Sources
• Make Generalizations
• Find common ideas across projects
• Synthesis of Data
• Use it as Thesis
• Defend belief based on series of compelling experiments
• Ask for validation
• Get Super basic functional baseline
• Write outline from common subsections between different papers
• Generalized form of different papers
• Sections for your projects
• Document research process and results
• 5 to 10 Pages ideal length
• Never Plagarize
• Omnigraph, inscape

Happy Learning, Reading and Writing!!!

Day #113 - Text Summarization Notes

This post is Summary for my reference on Text Summarization from Siraj Raval Session

Happy Learning!!!

Day #112 - Web Scraping

For topic modelling, had to scrap a few websites to obtain data for the same.

	#Python 3
	import httplib2
	import urllib.request
	import re
	from nltk.corpus import wordnet
	from bs4 import BeautifulSoup, SoupStrainer
	http = httplib2.Http()
	status, response = http.request('https://www.abc.com/india-car-news.htm/2')
	path_to_article = "D:\Scrap\Articles.txt"
	filehandle = open(path_to_article,'w')
	urltoadd = []
	for link in BeautifulSoup(response, 'html.parser',parseOnlyThese=SoupStrainer('a')):
	if link.has_attr('href'):
	if(('.htm' in link['href']) and ('http' not in link['href'])):
	print('https://www.abc.com'+link['href'])
	urltoadd.append('https://www.abc.com'+link['href'])
	for url in urltoadd:
	print('Article - Content')
	print(url)
	with urllib.request.urlopen(url) as url:
	page = url.read()
	page = str(page)
	soup = BeautifulSoup(page)
	txt = soup.body.getText()
	nstr = re.sub(r'[?|$|.|!]',r'',txt)
	nestr = re.sub(r'[^a-zA-Z0-9 ]',r'',nstr)
	print(nestr)
	words = nestr.split(" ")
	news = ""
	for word in words:
	if wordnet.synsets(word):
	news = news + word + ' '
	print(news)
	filehandle.write(news + "\n")
	filehandle.close()

view raw webscrapexample.py hosted with ❤ by GitHub

Happy Learning!!!