search

TomStog / Infrared-SpO2

The dataset used for the "A non-contact SpO2 estimation using video magnification and infrared data" publication
9 stars 0 forks source link

question #6

Closed gaosir123 closed 7 months ago

gaosir123 commented 7 months ago

Hello, I still have some questions. The 19 seconds in your video are divided equally. Are the blood oxygen saturation values ​​​​the same in these 19 seconds? If they are not the same value, how to deal with it?

TomStog commented 7 months ago

Of course SpO2 fluctuates but we made some assumptions in order to calculate it. One assumption was that the real SpO2 value was the value shown by the finger pulse oximeter. Another assumption was that the SpO2 doesn't fluctuate within our time window. Of course, the narrower the time window, the more accurate the prediction.

gaosir123 commented 7 months ago

If the videos used are of different lengths, will it affect the experiment?

TomStog commented 7 months ago

I don't know, different video lengths would mean different number of samples per training/testing. It feels like training a CNN on different size images. Can you give me an example of how different these video lengths are ?

gaosir123 commented 7 months ago

Then I would like to ask if your video is directly amplified by EVM and then extracted features, or does it need to be denoised and other operations?

TomStog commented 7 months ago

First, we apply motion attenuation and then motion magnification and feature extraction. There isn't any denoising step involved.

gaosir123 commented 7 months ago

How motion attenuation is achieved？

TomStog commented 7 months ago

I've uploaded the MATLAB code.You can find all the functions needed here: https://github.com/rgov/vidmag