Interpolation in 5 minutes
ฝัง
- เผยแพร่เมื่อ 15 ก.ค. 2024
- Equivalent to a 50 minute university lecture on convolution-based interpolation methods.
0:00 - intro
0:31 - 1D convolution
1:02 - linear interpolation with a hat filter
2:12 - deriving the sinc function
3:33 - ringing
4:19 - cubic and lanczos filters
4:51 - 2D interpolation filters
Graphics in 5 minutes is a series of cartoon-style videos that teach computer graphics in 10x less time. You can take the equivalent of a University level computer graphics course in just over two hours. The playlist is here:
• Graphics in 5 minutes ...
See here for more information: g5m.cs.washington.edu/
FINALLY!!!! THERE'S A CLEAR!! AND FREINDLY EXPLANATION OF WHAT I NEEDED TO COMPLETE MY CODE!!!!!!!!!!!!!!!!! THANK YOU SO MUCH!!!!!!!!!!!!!!!!!!
Please for the love of all that is holy keep making these videos. Why is it that almost all the best foundational math content used in Computer Vision applications I can find online comes from someone at The University of Washington? Y'all are awesome!
I really think the video is underestimated... it is clear, easy to understand, and comprehensive
What a beautifully simplified way to visualize the working of kernels and its relation to interpolation. WAW!
Omg. Years of formulae finally resolved.
This was a good one. Cheers!
1:16 multiplying the sample values(0.5) times the num of blue samples( i understand you gotta sum of their values)
0.5*1.2 = 0.6 and 0.5*2 = 1... then i need to sum the both results 1.6, and that would be the 50% from A way towards B.. and when they are .25 and .75 it goes like -->
1.2*0.25 + 2 *0.75 = 1.8 ..
Could you please explain how much space does it have to take from each calculation? Or How much jumps?
Also are the calculations from the filter similar to the following process?:
When the filter is between the two blue dots
Does it multiply the two values in the one column then add them together in the middle of the filter? Is this how it works?
I arrived at your video from the audio world, where there is a fellow who has made interpolation filters for reconstructing analog audio waveforms from digital samples. While an infinite-length sinc function can perfectly reconstruct a bandlimited signal from its samples, what is the interpolation error for finite-length sinc functions? This fellow (Rob Watts / Chord Electronics) uses upwards of a million (!!) taps of a sinc function to reconstruct the signal. With fewer taps (say, 100) would the estimation of inter sample peaks (like those encountered with drums) be shifted in time upon reconstruction?
hear me out. it'll be "Interpolation in 2.5 minutes" if you put the playback speed on 2x