|
From: Charles 'B. K. <kr...@ac...> - 2001-05-20 15:56:01
|
Libdv doesn't have a built-in scaler. It would be relatively easy to downscale to a thumbnail by doing DC-only decode (downsample by factor 8 in each direction). Otherwise, it would be a fair amount of work to get much performance benefit from an integrated scaler. One problem is that parsing takes up so much of the current time, and that does not change when you down sample. As a simple rule, every pixel in a block depends on every DCT coefficient. Since parsing currently accounts for about 50% of libdv time, the total reduction from scaling is limited to 50% of the reduction achieved, which will be exclusively in the backend. So if you can speed up the backend by 50%, your total speedup will be 25%. If anybody wants to do that, I'd suggest writing specialized versions of DCT, weight, and quant that work on 1/2 and 1/4 of the pixels. -- Buck ps Alternatively, there is a really interesting article in the most recent issue of IEEE Transactions on Circuits and Systems for Video Technology on doing scaling in the DCT domain. I don't know if their techniques would let you get speedups over non-scaled decode, but it looks very cool for arbitrary scaling in sofware, relative to other software techniques. "Dan Dennedy" <dde...@co...> writes: > Is there a way to tell the decoder to do some down-scaling while decoding? I > am interested in offering a smaller video window size in Kino to improve the > performance of dv decoding. Currently, in our preferences, I added a setting > for the quality index, and that helps. I would prefer that the decoder does > the scaling so it has less work to do. A downstream scaler would defeat the > purpose! > Thanks, > +-DRD-+ |