Visual Recognition
Instructor: Vinay P. Namboodiri
Lecture hours:
Tuesday 17:10 - 18:25 and Thursday 17:10 - 18:25
Venue: L19, LHC
Course Content
In this course we undertake a study of visual recognition from various aspects related to computer vision. Visual recognition the aim is to interpret semantic information from images. This is a task that humans excel in and the aim is to able to do so computationally. The kind of semantic information one seeks to obtain from images relate to the kind of entities present in an image (for instance, naming the kind of bird or type of vehicle present in an image). The challenge more generally can be thought of mathematically as one of learning a function Fw(x)-> y which takes a visual input x and generates a target output y by using a parameter vector w. Visual recognition is challenging due to the wide variety in the space of input x and the kind of outputs y. For instance, the images of even a restricted class of images such as faces exhibits lots of varieties due to factors such as pose, illumination, occlusion and orientation in addition to the inherent variety of human faces and therefore the task of recognizing faces is challenging. In this book we will consider the a variety of output tasks such as object recognition, object detection and object segmentation.
Current techniques based on deep learning are able to learn the above tasks but are able to do so assuming full supervision. However, obtaining such supervision for each task is challenging and not feasible always. Recently there has been interesting work towards solving these problems by reducing the amount of supervision available. This is done in various ways such as transfer learning, active learning, learning with weak supervision and unsupervised learning techniques. In the course, we aim to also consider such techniques that could be applicable for the various visual recognition tasks.
A brief outline of the topics to be covered in the course are as follows:
- Introduction to visual recognition and the various problems
- Instance Recognition
- Features for visual recognition
- Object Classification
- Classical to Deep learning
- Object Detection
- Object Segmentation
- Self Supervision
- Weak Supervision
- Domain Adaptation
- Unsupervised visual recognition
- Vision and Language
List of Teaching Assistants
- Aman Deep Singh
- Pravendra Singh
- Saket Jhunjhunwala
- Samik Some
- Siddharth Singla
- Utsav Singh
References
- Computer Vision: Algorithms and Applications by Richard Szeliski Available online
- Computer Vision: Models, Learning, and Inference by Simon J.D. Prince Available online
- Deep Learning by Ian Goodfellow, Yoshua Bengio and Aaron Courville Available online
- Computer Vision: A Modern Approach by Forsyth and Ponce Indian edition available
Course Discussion - Piazza
Link available over here
Assignment
Lecture Slides, notes and related reading