Lecture on Bag of Visual Words for Finding Similar Images Cyrill Stachniss, spring 2020 Note: Same lecture as for the 2020 C++ Course but without the project instructions
Жүктеу.....
Пікірлер: 12
@jeremiasgaia45733 жыл бұрын
Thanks a lot from a PhD student in Argentina. All of your videos are helping me so much in my thesis.
@fabryperot80813 жыл бұрын
Very insightful. Thanks Cyrill.
@first_m3m33 жыл бұрын
excellent material, highly recommended!
@comvnche2 жыл бұрын
Outstanding!
@amortalbeing2 жыл бұрын
This was really good . thanks a lot doc
@nicolasperez42922 жыл бұрын
if you normalize the vectors and take the Euclidean distances, you get the exact same query results as when you take the normalized vectors and measure the angles. it follows from the proof you wrote out. you can also visualize this intuitively by imagining the vectors as points on a unit sphere.
@amortalbeing
2 жыл бұрын
He showed the results, they are not 100% the same unless, after TF-IDF appliance, you normalize everything to 1, in which case you are 100% right.
@kdubovetskyi Жыл бұрын
50:28 -- you are telling that images 0 and 3 should be similar. I just can't understand why you make this assumption/statement. We don't have ground truths, and therefore how could we know, what image-distances should be: which of them are similar and which aren't. Could you please, explain why did you prefer the right matrix? And thank you so much for the lection and its 5-minutes summary. They are so insightful!
@nicolasperez42922 жыл бұрын
53:00 you forgot to include the factor of 2 in front of $x^{T} \cdot y$ on your second last line
@CyrillStachniss
2 жыл бұрын
Correct, the factor 2 was missing after the first "=", but the rest is correct. Thanks for spotting the typo!
@nicolasperez4292
2 жыл бұрын
@@CyrillStachniss no problem, thank you for the wonderful videos! do you have a video showing how the perspective projection is derived for homogeneous coordinates? this wikipedia page: en.wikipedia.org/wiki/3D_projection#Perspective_projection gives the formula for it using Euclidean coordinates.
Пікірлер: 12
Thanks a lot from a PhD student in Argentina. All of your videos are helping me so much in my thesis.
Very insightful. Thanks Cyrill.
excellent material, highly recommended!
Outstanding!
This was really good . thanks a lot doc
if you normalize the vectors and take the Euclidean distances, you get the exact same query results as when you take the normalized vectors and measure the angles. it follows from the proof you wrote out. you can also visualize this intuitively by imagining the vectors as points on a unit sphere.
@amortalbeing
2 жыл бұрын
He showed the results, they are not 100% the same unless, after TF-IDF appliance, you normalize everything to 1, in which case you are 100% right.
50:28 -- you are telling that images 0 and 3 should be similar. I just can't understand why you make this assumption/statement. We don't have ground truths, and therefore how could we know, what image-distances should be: which of them are similar and which aren't. Could you please, explain why did you prefer the right matrix? And thank you so much for the lection and its 5-minutes summary. They are so insightful!
53:00 you forgot to include the factor of 2 in front of $x^{T} \cdot y$ on your second last line
@CyrillStachniss
2 жыл бұрын
Correct, the factor 2 was missing after the first "=", but the rest is correct. Thanks for spotting the typo!
@nicolasperez4292
2 жыл бұрын
@@CyrillStachniss no problem, thank you for the wonderful videos! do you have a video showing how the perspective projection is derived for homogeneous coordinates? this wikipedia page: en.wikipedia.org/wiki/3D_projection#Perspective_projection gives the formula for it using Euclidean coordinates.
@sujandhali2483
7 ай бұрын
Read chapter 2 and 3 in multi view geometry book