Avisynth+ plugin modernization efforts

[Dogway]

Does anyone know what kind of convolution this theory applies to? It sounds to me as repair(mode=2) but I don't know, he also talks about a 2x2 window, how is that, should conv kernels be odd?

Originally Posted by madshi

First of all the contributors need to be sorted. The contributor which is nearest to the "center" (usually gets the highest weight) should be first in the list etc. The list should be sorted for distance to the center and not for weight. The sorted contributor list makes it easy to apply special processing to the "main contributors".
(...)
Hmmmmm... I didn't know the "Repair" filter. Seems that it uses the same concept, but it's not as well optimized for the specific case, I think. E.g. from what I've just read, "Repair" always compares 3x3 pixels. From my tests that's a good general purpose solution, but maybe not the best solution specifically for scaling. For upscaling I'm only comparing 2x2 pixels (those which are nearer to the "center"). For downscaling it can be more than 3x3 pixels, depending on the scale factor.

Or in short: The algorithm I described is custom tailored to "contributor" based resampling algorithms, by making use of the information in the contributor list. I think this should result in slightly superior quality because exactly those pixels are compared which are the main contributors to the final destination pixel. If we compare too many or not enough pixels, the algorithm should get a bit less effective (e.g. miss some ringing or misdetect ringing).

Here my attempt:

str = Format("y[-1,1] A^ y[0,1] B^ y[1,1] C^ y[-1,0] D^ y[1,0] F^ y[-1,-1] G^ y[0,-1] H^ y[1,-1] I^ x[0,0] J^"  \
            +" J 1 - K^ K A - abs K B - abs < A B ? L^ K C - abs K D - abs < C D ? M^ K F - abs K G - abs < F G ? N^ K H - abs K I - abs < H I ? O^" \
            +"                                         K L - abs K M - abs < L M ? P^ K N - abs K O - abs < N O ? Q^                               " \
            +"                                                   K P - abs K Q - abs < P Q ? R^                                                    " \
            +" J 1 + K^ K A - abs K B - abs < A B ? L^ K C - abs K D - abs < C D ? M^ K F - abs K G - abs < F G ? N^ K H - abs K I - abs < H I ? O^" \
            +"                                         K L - abs K M - abs < L M ? P^ K N - abs K O - abs < N O ? Q^                               " \
            +"                                                   K P - abs K Q - abs < P Q ? S^                                                    " \
            +"                                                             J R S clip                                                              " )

[Dogway]

Also is there any jpeg source filter for avs+ x64 like jpegsource that lets you load the image in original format (YUV) rather than convert it to RGB?

[zorr]

Originally Posted by Dogway

he also talks about a 2x2 window, how is that, should conv kernels be odd?

Not sure but 2x2 and scaling reminds me of bilinear interpolation.

[Reel.Deel]

Originally Posted by Dogway

Also is there any jpeg source filter for avs+ x64 like jpegsource that lets you load the image in original format (YUV) rather than convert it to RGB?

FFImageSource and LWLibavVideoSource can.

[kedautinh12]

SmoothUV2 now add back SSHiQ from old SmoothUV
https://github.com/Asd-g/AviSynth-SmoothUV2/releases

[GMJCZP]

Hello, the tritical's TUnsharp plugin does not work in HBD, is there a possibility to solve this situation?

[kedautinh12]

TUnsharp now don't support HBD. Need some one add support HBD

[Dogway]

I'm not sure that TemporalRepair is working as expected. modes 1, 2, and 3 barely do anything. I ported them to ex_repair and visually they clearly try to recover detail, unless I got the source code wrong.

RG_FORCEINLINE void get_lu_c(int& lower, int& upper, int src_prev, int src_curr, int src_next)
{
  auto max_np = std::max(src_next, src_prev);
  auto min_np = std::min(src_next, src_prev);
  upper = subs_c(max_np, src_curr);
  lower = subs_c(src_curr, min_np);
}

RG_FORCEINLINE int SmoothTRepair1_c(int dest, int lower, int upper, const int src_prev, const int src_curr, const int src_next)
{
  auto src_dest = dest;

  auto tmp_u = adds_c(upper, src_curr);
  auto tmp_l = subs_c(src_curr, lower);

  auto tmp_max = std::max(std::max(tmp_u, src_prev), src_next);
  auto tmp_min = std::min(std::min(tmp_l, src_prev), src_next);

  auto result = clip(src_dest, tmp_min, tmp_max);

  return result;
}

RG_FORCEINLINE int temporal_repair_processor_mode1_8_c(
  BYTE* dp,
  const BYTE* previous, const intptr_t pfpitch,
  const BYTE* sp, const intptr_t ofpitch,
  const BYTE* next, const intptr_t nfpitch
)
{
  int lowermax, uppermax;
  int lower, upper;
  get_lu_c(lowermax, uppermax, previous[-1 * pfpitch - 1], sp[-1 * ofpitch - 1], next[-1 * nfpitch - 1]);
  get_lu_c(lower, upper, previous[-1 * pfpitch + 0], sp[-1 * ofpitch + 0], next[-1 * nfpitch + 0]);
  uppermax = std::max(uppermax, upper);
  lowermax = std::max(lowermax, lower);
  get_lu_c(lower, upper, previous[-1 * pfpitch + 1], sp[-1 * ofpitch + 1], next[-1 * nfpitch + 1]);
  uppermax = std::max(uppermax, upper);
  lowermax = std::max(lowermax, lower);
  get_lu_c(lower, upper, previous[1 * pfpitch - 1], sp[1 * ofpitch - 1], next[1 * nfpitch - 1]);
  uppermax = std::max(uppermax, upper);
  lowermax = std::max(lowermax, lower);
  get_lu_c(lower, upper, previous[1 * pfpitch + 0], sp[1 * ofpitch + 0], next[1 * nfpitch + 0]);
  uppermax = std::max(uppermax, upper);
  lowermax = std::max(lowermax, lower);
  get_lu_c(lower, upper, previous[1 * pfpitch + 1], sp[1 * ofpitch + 1], next[1 * nfpitch + 1]);
  uppermax = std::max(uppermax, upper);
  lowermax = std::max(lowermax, lower);
  get_lu_c(lower, upper, previous[0 * pfpitch - 1], sp[0 * ofpitch - 1], next[0 * nfpitch - 1]);
  uppermax = std::max(uppermax, upper);
  lowermax = std::max(lowermax, lower);
  get_lu_c(lower, upper, previous[0 * pfpitch + 1], sp[0 * ofpitch + 1], next[0 * nfpitch + 1]);
  uppermax = std::max(uppermax, upper);
  lowermax = std::max(lowermax, lower);
  return SmoothTRepair1_c(dp[0], lowermax, uppermax, previous[0 * pfpitch + 0], sp[0 * ofpitch + 0], next[0 * nfpitch + 0]);

I understand that "sp[-1 * ofpitch - 1]" means "x[-1,-1]" otherwise my (unoptimized) expression is wrong:

y[-1,-1] z[-1,-1] max x[-1,-1] - U^
x[-1,-1] y[-1,-1] z[-1,-1] min - L^
y[-1,0] z[-1,0] max x[-1,0] - V^
x[-1,0] y[-1,0] z[-1,0] min - M^
U V max U^ L M max L^
y[-1,1] z[-1,1] max x[-1,1] - V^
x[-1,1] y[-1,1] z[-1,1] min - M^
U V max U^ L M max L^
y[1,-1] z[-1,-1] max x[1,-1] - V^
x[1,-1] y[-1,-1] z[1,-1] min - M^
U V max U^ L M max L^
y[1,0] z[1,0] max x[1,0] - V^
x[1,0] y[1,0] z[1,0] min - M^
U V max U^ L M max L^
y[1,1] z[1,1] max x[1,1] - V^
x[1,1] y[1,1] z[1,1] min - M^
U V max U^ L M max L^
y[0,-1] z[0,-1] max x[0,-1] - V^
x[0,-1] y[0,-1] z[0,-1] min - M^
U V max U^ L M max L^
y[0,1] z[0,1] max x[0,1] - V^
x[0,1] y[0,1] z[0,1] min - M^
U V max U^ L M max L^

x[0,0] U + y[0,0] max z[0,0] max
x[0,0] L - y[0,0] min z[0,0] min
a swap2 clip

[kedautinh12]

EEDI2CUDA ported from Vapoursynth
https://github.com/AmusementClub/Vap...ment-899069883

[kedautinh12]

EEDI2CUDA now have x86 ver
https://github.com/AmusementClub/Vap...uns/1135300628

[kedautinh12]

New fork frfun7 from pinterf, but still don't have HBD
https://github.com/pinterf/Frfun7/releases

[real.finder]

since there are no issues tab in https://github.com/pinterf/Average I will report it here

as said here, Average seems not work with avs+ native array


don't know if it can be updated without break compatibility

[pinterf]

New version: Average.
https://github.com/pinterf/Average/releases/tag/0.95

v0.95 (20211008)
- add AVX2 routines
- accept parameter as a nested array stuffed into the first parameter (AVS 3.7.1)
- pass over frame properties if any
- Source: update to VS2019, clang-cl option, gcc friendly
- add CMake build environment, linux build instructions

[real.finder]

Originally Posted by pinterf

New version: Average.
https://github.com/pinterf/Average/releases/tag/0.95

thanks! seems work fine

[Julek]

About Checkmate plugin: I wanted to point out that the wiki shows 5 as the default tthr2 but it seems that the correct is 0.