On Tue, Aug 3, 2021 at 12:48 AM Michael Niedermayer <mich...@niedermayer.cc> wrote:
> On Sun, Aug 01, 2021 at 04:08:17PM +0200, Nicolas George wrote: > > Linjie Fu (12021-08-01): > > > Checked swscale, it uses bicubic by default. > > > Scale filter sets the flags to bilinear if sws_scale_opt is not > initialized > > > by sws_dict. > > > > > > We should use the same default flags for both simple and complex filter > > > graphs, either from swscale, or from scale filter. > > > If prefer to keep it identical with swscale, we may modify the flags > in the > > > scale filter. > > > > Thanks. I also think that unless there is a good reason, we should rely > > on a single default, from the lowest level. So: > > > > - remove flags="bilinear" from vf_scale; > > > > - remove setting defaults in ffmpeg. > > > > Unless somebody can find a good reason for different defaults. > Michael, it is your commit, 8155bbc944c314fc202d31b0e4a3c77b8efc8dde, > > that introduced it. Any comments? > > > > Also, the current code in libswscale looks nonsensical: > > > > if (dstW < srcW && dstH < srcH) > > flags |= SWS_BICUBIC; > > else if (dstW > srcW && dstH > srcH) > > flags |= SWS_BICUBIC; > > else > > flags |= SWS_BICUBIC; > > > > It was originally 6b3ff6f91a535d6383f41ca7bdf760165dcb6015 with some > > logic, but it was cancelled by the merge commit > > 5710f55e490d0c3cf7348b3ca91c8c0fe4274be2. > > > "The default is left at bicubic until someone has compared the scalers > > properly speed and quality wise." > > > > But eight years later nobody did. > > In case someone tests this, it should test > 1. human subjective quality in addition to some objective quality > 2. scaling + compression > 3. various material, noisy and less crisp material may favor > different scalers than super flawless crisp images Just update the patch set as suggested and keep the default to "bicubic" for now, until we get a better choice, thx. - linjie _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".
