7. Constraints: Interpreting Line Drawings
ฝัง
- เผยแพร่เมื่อ 27 พ.ค. 2024
- MIT 6.034 Artificial Intelligence, Fall 2010
View the complete course: ocw.mit.edu/6-034F10
Instructor: Patrick Winston
How can we recognize the number of objects in a line drawing? We consider how Guzman, Huffman, and Waltz approached this problem. We then solve an example using a method based on constraint propagation, with a limited set of junction and line labels.
License: Creative Commons BY-NC-SA
More information at ocw.mit.edu/terms
More courses at ocw.mit.edu
This is priceless, a history lesson, a creation lesson, a process lesson, and deep insight into vision. MIT OCW you are angels.
Many love to Patrick Winston. He just passed away this Summer. Thanks for the great courses you given, it was an honor to be your student
So Sad to lose a great teacher and brilliant mind.
God bless Pat. May he rest in peace.
I want to clarify how to distinguish the concave, convex, and boundary lines.
1. If you see 2 faces on each side of a line, then the line is a concave or convex line, depending on the angle of the 2 faces: angle > 180 is convex, and angle < 180 is concave.
2. If you see 1 surface on a side of a line, on another side, it is space, air, or nothing, then, the line is a boundary line. The way to determine the direction of the arrow on the line: the surface(object) you see is always on the right side of the arrow.
i.e. if the arrow points to left in a horizontal boundary line, then the surface(object) is above the line;
if the arrow points to up in a vertical boundary line, then the surface(object) is on the right side of the line.
If I am not clear enough, leave a message here to ask me.
and how do we define the angle, is it the angle between surfaces on the side that we can see ? Because angles between two surfaces can be defined from either side and one is > 180 and the other is < 180
I got that but didnot understand at 21:30 why the line at the bottom is concave rather than a + at the bottom. Is it because of the angle one view that junction from?
very clear explanation, thx a lot
@@rakeshovr1 I think because we are looking at the line relative to the surface.
ty
What's amazing is that the professor knows the names of his students.
This blew my mind. Soooooo cool. The more I watch this course the more I want to binge watch it like an addictive TV series.
:P
So do. I wish I was able to finish it the same way I finish tv series.
R.I.P Patrick Winston
The prof's program said that the drawing at 47:55 was unambiguous and could represent only one scenario but i think 2 are possible.
1. A viewer watching a staircase normally such that goes from the left to the right
2. A viewer watching a staircase from below it such that it goes from left to right
Agreed to an extent. It is true that two percepts are possible but they cannot be possible at the same time so there is no ambiguity. They are both correct and contingent as long as you only focus on one. It would be ambiguous if somewhere in the middle the information doesn't correspond to either left-to-right or right-to-left.
That might also be the reason why the program starts in the bottom-left and continues with that orientation. Since there is no real ambiguity it may never switch to the other percept. We as humans can freely rotate & switch the figure. For the program to be able to do that it would need recurrent connections.
Hoffman is a genius. Once the 4 line types and 18 possible junctions are laid out, the rest is clear.
Whatever is in the cup, it's strong.
Ha
haha
Did not manage to figure out criteria according to which we classify a line as concave or convex.
Ghassen Smaoui As far as I understand, a concave line is the edge where a couple of walls converge, looking at it from the inside of your room, like the inner edge of a cube, and the convex would be the outside edge of the cube.Hope it helped.
Very helpful. I have a course that requires me to learn this and this made it very clear. Thanks.
I am finding this hard to grasp, is there any other resource one can suggest to understand this?
"Back when men were men" :D
Real men take at least 4 physics, and at least 4 math courses.
"Back when Men were Men " Preach it brother.
unbelievably smart.
This video is good thanks.
why cant exist a arrow with 2 bondaries and 1 concave? at 19.:18 you can see in this perspective
Elegant, yes. Simple no. Well, at least it wasn't simple for me. Still, this is MIT, and their students are the brightest of our country. Quite fascinating. I always wondered how machine vision worked.
Vision today doesn't look much like what you see here, though I'm really excited to find this lecture!
Lol yeah, I feel like it could've been explained simpler somehow though. All I got out of the lecture is that things can be distinguished through lines and vertexes and that those lines and vertexes can create constraints on what other lines and vertexes can be lol
awesome
Trihedral vertexes modeling
Awesome
21:02 How are 2 convex and one concave?
Shpuldnt they all be concave ?
i understood. nvm
how , can u help plz
@@HarshSingh-xe7mp he fused the paper box with the table making the 2 concaves. without the table would be 2 boundaries and 1 convex.
The most important question: what kind of chalk is that?
It's known as jumbo chalk or as railroad chalk.
Thank you for the reply, but I was meaning the brand, because it writes better than the chalk I have used at the 3 universities I have taught at.
If you listen closely at 46:42, somebody tries to cover up their fart by coughing and fails horrendously.
What's "stuff of the object" ?
I hate it when people use vague terms
Object on one side, air on the other...
The stuff can be roughly thought as the contents of the box... Arrows on the boundary should be such that the contents of the box are on the right of the arrow in the pointed direction..
6:50 you forgot 3 to 7
here is a little pdf ... drive.google.com/file/d/0B4s0uhYYRC1baVdfZG5NcGpJQWc/view?usp=sharing
showing a sketchup 3d the model @ 30:00 that could really exist
Of course it can, did you even listen to the professor?
If the lines are concave, is this really computer vision? The computer is seeing through the object at this point.
His explanation of what a boundary line is terrible. Could be improved by using his half cube and using his fingers to "walk" along an un-taped edge to better explain it. Other than that it's really good.
This is just a 4 state instead of 2 statisfiability problem. Normally SAT is boolean logic but if you image each clause as a vertex and each variable as a edge. Each edge can have 4 line notations. Much simpler approach, and his "depth" first search could have just been distribution of each clause. But the only academic accomplishment I have is a high school diploma -- something about his first lecture about give up your seat so someone else can be here.
How is he deciding the direction of boundaries ? XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXOOOOOOOOOOOOOOOOOahhhhhhhhhhhh
if you walk towards the direction of the arrow the right side of where you are walking should be the object not free space
The only i know is that if I can't visualise those line drawing into 3D object, I can never make a program that does that.
This is not the most complex lecture so far, but I can't see where it's going :/ I was 15 minutes in and I couldn't see any relation between the computer vision approach and what it has to do with constraints :/
All complex problems are understood by first studying the simplest cases. This is one of the simplest non-trivial vision problems one could imagine - how to interpret a 2d projection (the line drawing) of a highly constrained 3d world (made only of trihedral objects)
Are you serious? I think you did not understand the content...
From what I got, its different ways to make the computer determine 3d shapes in a 3d space.
Visual sensors on computers and human take in 2d information and infer a 3d environment. We do this through neurons that recognize edges specifically so a very natural approach to computer vision would be to understand edges and their relationships.
not really complex uptill 16 minutes
Dont know what you are talking about
Yeuteamo
How is this even important ?
I think this is redundant today(maybe wasnt 7 years ago)
just go away. nobody cares about your opinion here!
hi，I am confused about the direction of the arrow at 21:37,can you please help me to explain why
Dragos , I should say the same thing to you
Lei , It is very difficult to explain without a 3D model
He uses symmetry with the previous example to determine the direction of the arrow
Just watch the part where he uses the model to describe directions
But in reality , knowing this is not a big deal
Real AI stuff lies ahead
The details aren't important but the method is.
I thought MIT would have really good professors. I was so wrong.Even MIT is not spared from bad teaching.stuff of the object, octants ,so vague.
Quadrant 1: (x>0, y>0)
Quadrant 2: (x0)
Quadrant 3: (x
You're a fucking moron.
It's not that MIT have bad professors, most of them use vague terms because it's implicit that the student already know those concepts and has no need to waste time explaining that.
Its an abstraction dear..