Hands on Deep Learning for Computer Vision – Techniques for Image Segmentation.(6 hours workshop).
Submitted by usha rengaraju (@usharengaraju) on Sunday, 28 April 2019
Section: Workshops Technical level: Intermediate Session type: Workshop
Computer Vision has lots of applications including medical imaging, autonomous
vehicles, industrial inspection and augmented reality. Use of Deep Learning for
computer Vision can be categorized into multiple categories for both images and
videos – Classification, detection, segmentation & generation.
Having worked in Deep Learning with a focus on Computer Vision have come
across various challenges and learned best practices over a period
experimenting with cutting edge ideas. This workshop is for Data Scientists &
Computer Vision Engineers whose focus is deep learning. We will cover state of
the art architectures for Image Classification, Image Segmentation best practices in training and tuning deep neural networks. It will be hands on session in PyTorch v1.0.
The workshop takes a structured approach. First it covers basic techniques in
image processing and python for handling images and building Pytorch data
loaders. Introduce CNN based architectures for Image Classification (Resnet). Discuss ideas from cutting edge papers on Computer Vision and implement hands on in class best practices in on Image Classification and Semantic Segmentation architecture. Discuss how to train and tune deep neural networks.
Total Duration of Workshop – 6 hours
Part I – Duration: 1.5 hours
1. Introduction to Image Processing in open CV – Solve an end to end Computer Vision problem using OpenCV and discuss hands on concepts of Image Processing, Masking, Morphology operations, Contour Detection, Edge Detection etc
2. Introducing PyTorch – Tensor Operations and manipulations, Automatic Differentiation, Torch Vision and Data Loaders in PyTorch v1.0
3. Project 1 – creating data loaders in pytorch v.10 for Image Classification and Semantic Segmentation
Part II – Duration: 3 hours
1. Creating a Neural Network in PyTorch using torch.nn module, writing a training loop, optimizers and loss function.
2. CNN – Hands on implementation of CNN architecture (Resnet) for image classification, Data Augmentation.
3. Project 2– Time Series problem solving through Image Classification
4. Autoencoders – Hads on Implementation of Convolution Autoencoders
5. Semantic Segmentation – Discuss ideas of FCN paper- Skip Connections, Deconvolution, Removal of fully connected layers
6. Implement U-Net based architecture in PyTorch v1.0
7. Project 3– Semantic Segmentation to generate mask on medical image dataset using U-Net architecture
Part III – Duration: 1.5 hours
1. Practical advice and best practice to train deep neural network
2. Tuning hyperparameters – discriminative Learning rate, one cycle policy, mixed precision etc. Show ipython notebooks in detail as to how to implement them.
3. Practical implementation of best practices in training deep neural networks – Initialization, batchnorm, early stopping, minibatch training, callbacks and hooks in training loop
4. Revisit above hands on projects and experiment with the best practices learnt in the class
Target Audience –
• Computer Vision Engineers
• ML/Deep Learning Engineers/Architects
• Data Scientist – who wants to pick up Computer Vision
• Deep Learning practitioners in general
• PyTorch developers
Who should not attend this workshop –
Practitioners who are not interested in ML/DL/Computer Vision should not attend this workshop.
Software Requirements – Laptops with internet connectivity, PyTorch v1.0, Python v3.6, OpenCV to installed in the machine, good to have access to GPU else will use google colab to run hands on experiments in class.
Currently authoring a book on “Deep Learning with PyTorch” with Apress. Has overall around 12 years of experience working in Data Science. Currently leads the Deep Learning team at Dell IT working out of CFO office (Tom Sweet) on mandate to transform Cash Application towards making back office Finance frictionless- Automating the data entry through Deep Learning. Set up a deep learning team in Dell from scratch. Previously head of Advanced Analytics for Infocepts build the analytics team from scratch. He has vast experience working with terabyte scale data and has been developing solutions in AI space. Have worked for consulting companies in India (PwC &vTeradata Professional Services) delivering data science solutions. Has strong experience in building data science practice from scratch and has diverse skills ranging from data engineering, bridge between business and technology, Information Modelling, data science. Has experience working with Banking and Insurance, FMCG & Retail. His research interests include Deep Learning, Computer Vision, and Information Retrieval. . Has earlier presented in World Machine Learning summit India organized by 1.21GWs, December 2018 on Semantic Segmentation and a workshop on Medical Imaging and gave talk at Microstrategy world on Advanced Analytics in 2016.
1. Was selected in Microsoft Academic Project Program (MSAPP) by MSDN for my work on HR Analytics for SME IT, 2006.
2. Employee Excellence Award - TCS, 2009
3. Head of Asia for Citi Genesis Project- Recognition Award for Data Analysis, 2011
4. Awarded from Dell CFO (Tom Sweet) on automating the Data Entry work using AI, 2018.
5. Build AI practice by getting funding approved on the basis on the AI work we did for them.
6. Designed largest banking data warehouse for Citi- Genesis Program for Wholesale and Retail Banking products, 2011.
I am a polymath and unicorn data scientist with strong foundations in Economics, Finance, Business Foundations, Business Analytics and Psychology. I specialize in Probabilistic Graphical Models, Machine Learning and Deep Learning. I have completed Financial Engineering and Risk Management program from Columbia University with top honors, micromasters in Marketing Analytics from UC Berkeley and statistical analysis in Life Sciences specialization from Harvard. I am chapter lead/Co-Organizer of Women in Machine Learning and Data Science Bengaluru Chapter and Core organizing team member at WIDS Bengaluru .I have around 6 years of technical experience working in various companies like Infosys, Temenos, NeoEYED and Mysuru Consulting Group. I am part of dedicated group of experts and enthusiasts who explore Coursera courses before they open to the public, an ambassador at AIMed (an initiative which brings together physicians and AI experts), part time Data science instructor, mentor at GLAD (gladmentorship.com), mentor at JobsForHer and volunteer at Statistics without Borders. I developed the course curriculum for Probabilistic Graphical Models @ Upgrad which is taught by Professor Srinivasa Raghavan from IIIT Bangalore.