VideoFrame.to_ndarray performance

[YAGregor]

to_ndarray this function call VideoFrame.format but we know that, to achievebest performance (when reformat multiple video frame), it's better to use one Reformatter. In my use case, I need low latency (< 4ms), but if I use to_ndarray. it create many Plane VideoFormat, etc, and them reference to each other which make them cannot free by ref count, and because I stored the ndarray for a while, finally thousands of Plane, VideoFormat will go to generation 2 in GC, cause a generation 2 gc cost hundreds ms.

is there any way to avoid this (without copy memory) or ideas to improve code?

[WyattBlue]

We've made PRs that make progress on this. Performance in general can be discussed here: #2009