[Pycon] [new paper] "Kajal Puri" - Object detection and Human recognition with YOLO in Python

[info a pycon.it]

Title: Object detection and Human recognition with YOLO in Python
Duration: 45 (includes Q&A)
Q&A Session: 0
Language: en
Type: Talk

Abstract: New research papers for object detection coming out every other day made really difficult to decide on one algorithm.I chose YOLO after analysis and OH BOY, how much i love my decision! Here,I’ll try to give an intuitive explanation behind the choice of YOLO, challenges faced and how to overcome them.

The structure of my talk will follow the following timeline: 

(0-10 minutes) - Introduction 
1. Difference between the problem statement of Human/Object Detection and Recognition. 2. Demonstration of how human tracking in a video plays a pivotal role in human counting as well as recognition over various frames.

( 10-25 minutes) - Contradistinction of Human recognition and Object detection Algorithms 
1. Very brief introduction to YOLO/DarkNet Model. 2. Comparison of YOLO2 with OpenCV’s Haar Cascade Classifier for real-time human detection. 3. Things to take care while deploying the YOLO2 with Python.

(25-30 minutes) - How to train your OWN YOLO2 and use the trained weights in Python 1. How much data a.k.a diverse data is sufficient to train your own classifier? 
2. Effect of not using pre-trained YOLO model on scalability 
3. How much accuracy is enough to finalise your model. 
4. Importance of Hyper-parameter tuning while training.

(30-32 minutes) - Video File Demo 
1. Working demonstration of Human detection on a video as well as on an image. 
2. Get the output and save it in a csv file with the particular time/frame slot.

(32-40 minutes) - Use-Cases/Applications 
1. Generate Heat-Map in a mall/shop to analyse the crowd presence on different days/timings during the week. 
2. For surveillance purposes in offices, banks etc. 
3. Expansion on advanced level leads to provide significant help in motion detection. 
4. Gender Classification

(40-45 minutes) - Interactive Q&A session with listeners.


Tags: [u'image-processing', u'ComputerVision', u'Data Mining', u'Machine Learning']