mara004
commented
10 months ago Big +1 for merging from a pypdfium2 dev POV.
Note, a major problem with the transfer is that, if receiving code is slower than the producing pool, images silently queue up in memory, which can cause enormously high memory loads. (PNG saving particularly exhibits the problem. JPG is usually fast enough to somewhat keep up with the pool.)
Addresses https://github.com/facebookresearch/nougat/issues/110
This removes multi-processing, which may or may not lead to a performance hit when loading documents with many pages. On the one hand, with this modification, nougat now uses a single process for sequential rendering of all PDF pages. On the other, it now avoids copying bitmaps across processes.
The benefit is that this makes it possible to implement nougat in already parallelised workflows (e.g., when reading multiple PDF files, each in a separate process)