The Basics about Bundle Adjustment Cyrill Stachniss, Fall 2020 This is Part 1 on a lecture on Bundle Adjustment, see Part 2 here: • The Numerics of Bundle...
A saviour video for my PhD study, cannot imagine there can be such a nice professor who is willing to devote so much effort and time to make everything so clear for free!
Too good - I wish I had studied your videos 10 years ago when I was starting out. Somehow, the books don't give an intuitive picture making this a much more difficult area to approach than it should be. Prof. Stachniss, you should write a book with some good pen and paper and programming exercises. Forstner is probably the best right now. (I work in self driving cars, on BEV modelling, and LOVE this subject).
Thank you so much for your content! My whole postgraduate research lab are huge fans of yours. You're getting us all through our dissertation-only Masters. Thank you Prof. Stachniss
Thank you for the great content Mr Stachniss. It really helps to clarify and make this knowledge more accessible. You deserve a million subscribers at least.
Can someone explain how we get 10M scale parameters, I thought scale factor would be defined for a camera pair and not be different for every point in the pair of images..
I think it is because we are using homogeneous coordinates. For each 3D point you get only a homogeneous representation after applying projection. The scale in this homogeneous representation (the z coordinate) can be (and I guess usually will be) different for each point. Hence the scale parameter lambda varies across points
Thanks for your great leacture! I have some question. In 1:02:15, What kind of variables when you mentioned variance factor? Does it include all kinds of unknown parameters? Thanks,
Hi, I wanted to ask you something about the scale factor. You avoid using this factor when you move from homogeneous world to euclidean world. Afterwards, you say the result of bundle adjustment does not contain any information about absolute scale. I wonder if this happens because of the change from homogeneous to euclidean world. Thank very much in advance for this amazing work!
The answer is no - these are two different (unrelated) scales. Even if you do BA in homogenous coords (i.e. do not get rid of lambdas), you will not be able to recover the absolute scale of the object in the world.