1
00:00:00,250 --> 00:00:03,337
Once you start looking under the hood, you'll get some sense of the lay of the

2
00:00:03,337 --> 00:00:06,769
land with matrices and how they relate to transforms. Here's a map of the sort

3
00:00:06,769 --> 00:00:11,246
of things you'll find. The upper left area of the matrix is where rotations and

4
00:00:11,246 --> 00:00:15,963
scales show up. If a transform changes only this area of the matrix, it's called

5
00:00:15,963 --> 00:00:19,985
a linear transformation. I'm not going to spend time on the formal definition of

6
00:00:19,985 --> 00:00:24,214
this term. The additional course materials include resources for more on the

7
00:00:24,214 --> 00:00:28,010
underlying math. The definition is fairly basic stuff about how addition and

8
00:00:28,010 --> 00:00:31,894
multiplication are preserved. But doesn't have much effect on how we actually do

9
00:00:31,894 --> 00:00:36,060
computer graphics. The upper right is where the translations accumulate. These

10
00:00:36,060 --> 00:00:39,544
translation values will get effected by multiplication with other matrices of

11
00:00:39,544 --> 00:00:43,896
course. Translations only effect points since vectors have zero for their fourth

12
00:00:43,896 --> 00:00:49,008
coordinate. 3JS provides a function called decompose to extract the translation,

13
00:00:49,008 --> 00:00:53,871
rotation and scale factors from a matrix. The translation and scale factors come

14
00:00:53,871 --> 00:00:58,557
back as vectors, as you might expect. The rotation comes back as a quaternion.

15
00:00:58,557 --> 00:01:02,874
Something we'll talk about when we get to animation. The short version is that a

16
00:01:02,874 --> 00:01:07,122
quaternion is a compact way to store the axis and angle of rotation for a

17
00:01:07,122 --> 00:01:11,707
rotation matrix. One useful property quarternions have is that you can easily

18
00:01:11,707 --> 00:01:15,861
interpolate between them, which is useful for interpolating between two

19
00:01:15,861 --> 00:01:20,100
different orientation. Notice that the bottom row is always 0, 0, 0 1. The

20
00:01:20,100 --> 00:01:24,597
transforms we covered here are called affine transforms. Parallel lines stay

21
00:01:24,597 --> 00:01:29,703
parallel when an affine transform is applied. In modeling you'll essentially

22
00:01:29,703 --> 00:01:34,304
always use affine transforms. So we never change this last row. Since GPUs are

23
00:01:34,304 --> 00:01:39,104
tuned to use four by four matrices Most of us just use four by fours everywhere

24
00:01:39,104 --> 00:01:43,579
for simplicites sake. When we discuss perspective cameras, we'll set the values

25
00:01:43,579 --> 00:01:48,230
in this last row. We'll then be using a projective transform. With affine

26
00:01:48,230 --> 00:01:52,523
transforms, when a points coordinates are multipied by the matrix. The fourth

27
00:01:52,523 --> 00:01:56,452
coordinate starts out as one and ends up as one. This last row in a projection

28
00:01:56,452 --> 00:02:00,324
matrix modifies that fourth coordinate to be something other than one. What that

29
00:02:00,324 --> 00:02:03,623
means is something we'll leave til a later lesson.