Scene 1

Scene 2

CUDA ( Compute Unified Device Architecture ) is a hardware architecture and programming model for parallel processing of Graphics Processing Units (GPUs).

A circuit board

Description automatically generated

Low-level programming to achieve very high performance

Extension of C/C++ Extension of  Java, Fortran, Python, and  Matlab Accelerated libraries as OpenCV, TensorFlow, and  Pytorch

Applications range from  biomedical imaging, oil and natural resource exploration, computer vision, bioinformatics, life science and artificial intelligence.

Scene 3

Direct use  of CUDA for a simple yet common problem like image interpolation.  Getting familiar with the functions running on the GPU, the  kernel  functions.

Topics covered : Nearest-neighbor interpolation Linear and bilinear interpolation CUDA texture memory Texture filtering Nearest-neighbor and linear interpolations of a PGM image Cubic B-spline interpolation Bicubic B-spline interpolation of a PGM image Texture lookup Catmull-Rom interpolation

Scene 4

Fundamentals of C/C++ and CUDA programming Basic elements of calculus, especially function approximation.  Suggested introductory reading on CUDA :  J. Sanders, E.  Kandrot , "CUDA by Example: an Introduction to General-Purpose GPU Programming", Addison-Wesley Suggested advanced book on CUDA : D.B. Kirk, W.-m.W. Hwu,  Programming Massively Parallel Processors: A Hands-on Approach ", Morgan-Kaufmann GitHub link for all the code files : https://github.com/CIuliusC/CUDA_Book/tree/master/Chapter%2001.

Scene 5

Diagram

Description automatically generated

Reconstruct the values of a function by properly combining its samples

Interpolation points : points where the function needs to be reconstructed,  located in the region covered by the function samples

Extrapolation : reconstructing a function  in the region outside that covered by the samples . Not dealt with in this Course.

Scene 6

Green curve : a certain function  Red dots : its samples   Yellow crosses : interpolation points

Interpolation techniques : use uniform or non-uniform samples

In this Course, we will deal with equally  spaced samples and consider most common interpolation techniques.

Scene 7

Prometheus
127 high (10 m/px) and medium (180 m/px) resolution in C21 context (1.3 km/px)
5km

Scene 8

A large air plane on a runway at an airport

Description automatically generated

Scene 9

A close up of a person

Description automatically generated

Scene 10

Illustrate how a particular GPU feature like texture memory enables performing, directly in hardware, basic interpolations as nearest-neighbor or linear.  Discuss how more complex interpolations, like cubic B- spline , may benefit from texture memory.

Writing CUDA kernels for interpolation

Page 1 (0s)

Page 2 (7s)

Page 3 (32s)

Page 4 (54s)

Page 5 (1m 20s)

Page 6 (1m 39s)

Page 7 (1m 54s)

Page 8 (2m 6s)

Page 9 (2m 14s)

Page 10 (2m 22s)