search

Stergrim / PIV-Particle-Image-Velocimetry

В репозитории представлена реализация метода кросскорреляции для определения локальных смещений оптического потока.
1 stars 1 forks source link

contributing back to OpenPIV #1

Open alexlib opened 5 days ago

alexlib commented 5 days ago

Dear @Stergrim, This is one of the most interesting derivatives of OpenPIV. How could you extend OpenPIV with your developments?

Stergrim commented 4 days ago

Dear @alexlib, to answer your question briefly, there is no way, in any way, I could expand OpenPIV with my own developments. I will answer in more detail below. My PIV project is just a modified version of what I did at university. I'm not interested in that right now. I gave up my scientific work. I see further development of the PIV field only with the use of neural networks, since they have higher processing speed and accuracy (for example, PIV-LiteFlowNet-en). The disadvantage of neural networks is their narrow focus, that is, the lack of flexible configuration for various cases. But to solve this problem and develop neural networks, I do not have the necessary knowledge and skills. In my master's thesis, there is a comparison of a neural network with a self-written PIV method. If you are interested in this, I can tell you a little more about what disadvantages I have found in neural networks and how I see their solution. Have a nice day.

alexlib commented 4 days ago

Thanks for a prompt answer. I would appreciate a chat about the comparison and your view on neural networks and PIV

shall we get on zoom? could you share your comparison ? I don't want to run all the NN again and would like to learn from your experience.

Alex alex dot liberzon at gmail dot com

Stergrim commented 4 days ago

I think it makes no sense to switch to zoom, since my knowledge of English is not enough for speaking. I will prepare an answer comparing neural networks and PIV, but this experience is limited to 2021. A detailed answer will take some time, more than one day. If you want to get acquainted with my research now, then my master's thesis is in the attachment. Paragraph 1.3 is an overview of articles on machine learning, and paragraphs 2.3 and 2.4 are the application and study of the limitations of neural networks. https://github.com/Stergrim/Recalibration-of-a-stereo-pair-based-on-a-reprojection-error/blob/main/demos/DiplomMaster.pdf