Object Detection with PyTorch

From simplest models to current State of The Art

Our main goal is to give you a deep understanding of ideas and problems that stand behind the Object Detection task by walk you through the history of development with the use of practical lectures.

About this Course

Object detection is a key task in Computer Vision. It has a great history of development, back from times when computers were too heavy to carry them around. Object Detection is an evolutional step for a wide range of applications such as military applications, urban planning, and environmental management. Nevertheless, it's a challenging task due to the different scales and appearances of the objects.

This course was designed as a next step of learning for people who already got basic knowledge from Coursera, Udemy, edX, etc and for people who want to expand their expertise in another area of Machine Learning. We are bridging the gap between theory and practical knowledge so you will be ready for real-life projects, master computer vision, and image processing and get a better job within months of the course completion.

It is created by developers to developers and provides a deep understanding of the object detection tasks in the computer vision field. Topics of the course will guide you through the path of developing modern object detection algorithms and models, so you will learn to apply deep learning architectures to computer vision tasks. Discover how to proceed in researching this area and create a more accurate solution. Topics of the course include simple localization models (based on coordinates and mask), single shoot networks (Yolo, SSD) and regional proposal networks (Faster RCNN, Mask RCNN).

We recommend our Object Detection Course as the perfect point to advance your deep learning education.

aspect_ratioFormat: Runnable notebooks with theory

trending_upEnvironment: Fast start with free Google Colab Environment. Ram: 12GB, Disk: up to 350GB, GPU: Nvidia T4 16GB

chatInteraction: Google Classrom + GitHub

access_time Time: Aprox. 4 weeks to complete

language Language: English + (Ukrainian, Russian in chat)

notifications_active Next session: 14th of December

Syllabus - What you will learn from this course

FREE

Classical computer vision

Road Traffic Counting with OpenCV
access_time4 hours

This lecture will teach you one of the classical ways of detecting objects without using neural networks and also will teach how to build an image processing pipeline for the task - Road Traffic Analytics.

play_circle_filledRun Notebook

WEEK 1

Introduction to PyTorch and Localization

PyTorch Basics
access_time2 h

PyTorch is currently the Top1 ML framework among researchers. This lecture will give you an understanding of building blocks of PyToch and shows how to use them to build training and inference pipelines.

Single Object Localization
access_timeTheory: 3 h Practice: 2 h

Object detection can be hard if start learning it from the end. This lecture will teach make intro in detection task through simple localization models with using coordinates and masks.

WEEK 2

Single Shot Networks

You Only Look Once (YoloV1)
access_timeTheory: 6 h Practice: 6 h

YOLO - fist network that was aimed at fast inference. In this lecture we will get a deep understanding of its internals and why it was built in such a way and what problems developers tried to solve.

Single Shoot Detector (SSD300)
access_timeTheory: 6 h Practice: 1 d

SSD - doing the same task as YOLO but was developed few months later by another researchers and it more robust then YoloV1. We will learn internals of this network and understand in what things it differs from YoloV1 and why.

WEEK 3

Regional Proposal Networks

Fast R-CNN
access_timeTheory: 8 h Practice: 3 d

Regional Proposal Networks - came to us from classical computer vision and was updated with convolutional networks. In this lecture we will learn the history of such networks and put our hands in internals of Faster RCNN.

Mask R-CNN
access_timeTheory: 8 h Practice: 1.5 d

Object detection evolves every day and today is a good thing to create multi-task networks and not only because then can solve few tasks in the same time, but also because they achive much higher accuracy then ever. In this lecture we take a look on the internals of curent state-of-the-art algorithm - Mask RCNN.

Instructor

Andrey Nikishaev

Mathematical department, Taras Shevchenko National University of Kyiv
ML Tech Lead at 20KL

Entrepreneur and Software Engineer with 15+ years of experience. Specialist in HL+HA and BigData Systems. Was working with Top10 IT product companies. The last 3 years mostly engaged in ML projects. Currently developing a Real-Time gaming recommendation engine heavily based on Computer Vision.

Reviews

Register to Course

By enrolling you agree with
Distance Learning service Agreement and Privacy Policy