Open Yzichen opened 4 years ago
All the math should be explained more or less clearly in the paper and the references. Unfortunately I don´t have time to spend on this, but I am sure you can solve your issues by reading scene flow papers or general computer vision ref. books.
Hello, the scene flow I know now has application in Moving object detection、segmentation、 SLAM、Autonomous vehicles、Depth image super-resolution,Where else do you think it can be applied ,especially that is not easy to think of. In addition, can your MC-flow be open source?
I cannot release MC-Flow because it's old & untidy code, I don't have time to clean it and I don't gain anything from publishing it.
Regarding applications: one you have not mentioned is non-rigid tracking / reconstruction. If you check papers like DynamicFusion you'll see they estimate a volumetric scene flow (I think they called it warping field). Another application could be temporal frame interpolation in videos, i.e. use scene flow to generate inter-frames to create something like a slow-motion video.
What should I do if I want to use scene flow for people counting?
I have a question, why not do scene flow estimation after refining the background? But at the same time scene flow estimation and refinement odometer?
Mariano Jaimez Tarifa notifications@github.com 於 2019年10月19日 週六 上午6:51寫道:
I cannot release MC-Flow because it's old & untidy code, I don't have time to clean it and I don't gain anything from publishing it.
Regarding applications: one you have not mentioned is non-rigid tracking / reconstruction. If you check papers like DynamicFusion you'll see they estimate a volumetric scene flow (I think they called it warping field). Another application could be temporal frame interpolation in videos, i.e. use scene flow to generate inter-frames to create something like a slow-motion video.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/MarianoJT88/PD-Flow/issues/11?email_source=notifications&email_token=ANALTXOLKKJO5DWZPVPU4O3QPI4Y5A5CNFSM4IYTA3X2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEBWKQNA#issuecomment-543991860, or unsubscribe https://github.com/notifications/unsubscribe-auth/ANALTXLQMJNVR4RP2LM3DZTQPI4Y5ANCNFSM4IYTA3XQ .
I tested your program, the effect is very good, you are now my idol.However, in the process of learning optical flow and scene flow estimation, I always encounter some formulas that cannot be derived, and some ideas of the author are not quite understood. Should I add some mathematical foundation, such as convex optimization?Nowadays, many people estimate scene flow through deep learning without too much mathematical derivation. What do you think?