Thanks to @garagecoder’s help, I had upgraded Reverse Engineer Gradient Map/rep_regm filter. It is now completely usable in cli, and gui as well.
I added multi-threading support for Chirikov-Taylor GUI/CLI filter. 6 times faster than serial processing in my computer. Also, added Alpha Support for Grayscale Mode in Chirikov-Taylor GUI filter.
Rewritten the rep_sptbwgp again. It’s faster. However, I just don’t get why there’s a speed difference in horizontal and vertical processing regardless of using a rotated image or not. Nothing makes sense about it. Permute doesn’t really solve it seems.
Now, I had managed to optimize rep_axis_streak_color again. I think I am done for the new year upgrade to my filters. I still however would like to be able to solve the issue of why vertical processing is slower than horizontal processing in rep_sptbwgp, and never did, and this is the only case where this bug shows up in terms of gmic processing or I am missing something. It doesn’t show up in other gmic scripts that utilize processing along row/columns, just this one which points to “missing something”, and nothing seem to indicate what it is.
Roadmap for 2021:
-
Refactor Naeddyar mapmaking script to more suitable version. He wrote scripts with little knowledge of gmic scripting. So, hence the need of refactoring. ← I’m here now.
-
Refactor few of my script and adding new feature.
-
New filters only after the refactor. I won’t be refactoring my script for optimization, etc.
Looks like I may work on Perspective Streak, and addressing problems. However, there is a issue I ran into with writing the Axis Streak Part. The axis streak editing part is meant to address a shortcoming with not being able to preserve image structure.
The problem: When setting _add2new to 1, the output is not as expected because if I were to blend the output with the old image, then areas where there are already non-zero transparent pixel would be off.
How do I address it with this code?
EDIT: I had addressed it. Realized I can just insert temp vector to new image before calculating newcol and newalp.
New Axis Streak
#@cli rep_axis_streak_color: orientation,direction,_alpha_exponential_factor,_maxval>0,_add2new={ 0=add2old | 1=add2new },_cmykmode={ 0=non-cmyk | 1=cmyka_mode }
#@cli : Streaks colored pixels taking into account of opacity.
#@cli : '_alpha_exponential_factor' is used to manipulate the alpha mixing within pixels. The more power that is assigned to the alpha, the more mixing there would be.
#@cli : '_maxval' divides the alpha channel internally to normalize ranges to 0-1. A error will appear if not normalized. By default, it used the max alpha channel value.
#@cli : '_add2new' defines whether to streak data on old image or to streak pixel data onto a new image. Used for Perspective Streak.
#@cli : '_cmykmode' is only used in case of using only cmyk images. Not needed in normal cases at all.
#@cli : Default values: '_alpha_exponential_factor=0','_maxalp=n/a','_cmykmode=0'
#@cli :
#@cli : Author: Reptorian.
skip ${3=0},${4=},${5=0},${6=0}
tcr=3
if $6 tcr+=1 fi
if narg($4) if $4==0 return fi fi
repeat $! l[$>]
if s==1||s==$tcr break fi
vv=0
repeat s
sh $>
vv+={iv#-1}
rm.
done
if !$vv break fi
sh. 0,{s-2}
sh.. {s}
f.. i#-1?I
if narg($4) alp={abs($4)}
else alp={iM#-1}
fi
/. $alp
if iM#-1>1||im#-1<0 error alpval(valid)==F fi
if $3<=-1 exp_f=-{1-10^-8} else exp_f=$3 fi
f. i^(1+$exp_f)
if $5 {w#0},{h#0},{d#0},{s#0}
col_info=-3
alp_info=-2
targ_info=-1
else
col_info=-2
alp_info=-1
targ_info=0
fi
if $1
outdata_dim={w},1,{d},{s#0}
outdata_coords=x,yy
if $2 direction=yy=hh-1,yy>=0,yy--
else direction=yy=0,yy<hh,yy++
fi
else
outdata_dim=1,{h},{d},{s#0}
outdata_coords=xx,y
if $2 direction=xx=0,xx<ww,xx++
else direction=xx=ww-1,xx>=0,xx--
fi
fi
$outdata_dim,":begin(
const ww=w#0;
const hh=h#0;
);
start_val=1;
for("$direction",
start_val?(
start_val=0;
col=I(#"$col_info","$outdata_coords",z);
alp=i(#"$alp_info","$outdata_coords",z,0);
temp=[col,alp];
):(
newcol=I(#"$col_info","$outdata_coords",z);
newalp=i(#"$alp_info","$outdata_coords",z,0);
newinfo=[newcol,newalp];
!newalp?(I(#"$targ_info","$outdata_coords",z)=temp;):
newalp==1?(temp=newinfo;
):(
col=newcol*newalp+(1-newalp)*col;
alp=alp+newalp*(1-alp);
temp=[col,alp];
I(#"$targ_info","$outdata_coords",z)=temp;
);
);
);"
if $5 rm[^-2]
else rm[1-3]
fi
sh. {s-1} *. $alp rm.
endl done
Looks like upgrading Perspective Streak is going to take a while. That being said, I added 3 new formulas to Thorn Fractal:
- Chaotic Hooks Unearthing
- Sinusoidal Liquid
- Cosinusoidal Liquid
Should be here after 1 hour.
Another 3 Formulas has been added to Thorn Fractal.
Side note: I realized with what @garagecoder shown me that filling outside images with fill is possible, it is actually feasible to do multithreaded version of Diffusion Limited Aggregation.
EDIT: I tried, didn’t really saved time. That means I’m done with current filters except for Perspective Streak.
1/11/2021:
Okay, now I have a new update. It seems that I am a bit closer to upgrading Perspective Streak. I’m working on transformation now. Now recpoltrans -3 to recpoltrans -2 takes .32 s as opposed to 1.5 s. More than 4.6 times faster. It shows promise at preserving more details too. After transformation, I would need to add more options to axis_streak_color just for the fact that it is crucial to make it work correctly with it.
1/14/2021:
Wow! Now I think I solved recpoltrans! The new changes is by far more faster. From 1.5 s to .194 s. It is also more accurate. I only worked on if -3, and 2 part of rep_recpoltrans since they matter a lot more than the others as they’re both designed for precision. The others if in rep_recpoltrans don’t really matter too much for me to fix them. In addition to these changes, I’m closer to fixing a bug in perspective streak!
See here.
Left: Original Middle: recpoltrans -3 to recpoltrans 2 Right: Xor Analysis
[gmic]-2./ Elapsed time: 0.19 s.
Also, saving code here:
Code of new_reprecpoltrans
new_rep_recpoltrans:
skip ${1=0},${2=0},${3=0}
if $1<-1||$1>1 error "($1>=-1&&$1<=1)=0" fi
if $2<-1||$2>1 error "($2>=-1&&$2<=1)=0" fi
repeat $! l[$>]
ov=${-rep_2dcr}
if $3==-3
maxlength={max(w,h)*1.5}
perimeter={(w+h)*2}
{$perimeter},{$maxlength},{d},{s},":begin(
surface=expr('begin(const dpi=2*pi;);(x/w)*dpi;',w);
const dpi=pi*2;
const ww=w#0-1;
const hh=h#0-1;
const ihh=h-1;
const point_x=(($1*-1)*.5+.5)*ww;
const point_y=($2*.5+.5)*hh;
const inv_point_x=ww-point_x;
const inv_point_y=hh-point_y;
const cut_ang_s0=abs(atan2(inv_point_y,inv_point_x));
const cut_ang_s1=pi-abs(atan2(inv_point_y,point_x));
const cut_ang_s2=pi+abs(atan2(point_y,point_x));
const cut_ang_s3=dpi-abs(atan2(point_y,inv_point_x));
distanceaway=[ww-point_x,hh-point_y,point_x,point_y];
);
surface_angle=surface[x];
surface_angle>cut_ang_s0&&surface_angle<=cut_ang_s1?side=1:
surface_angle>cut_ang_s1&&surface_angle<=cut_ang_s2?side=2:
surface_angle>cut_ang_s2&&surface_angle<=cut_ang_s3?side=3:
side=0;
mdist=abs(side%2?1/sin(surface_angle):1/cos(surface_angle));
dix=(point_x+cos(surface_angle)*distanceaway[side]*mdist*y/ihh);
diy=(point_y+sin(surface_angle)*distanceaway[side]*mdist*y/ihh);
I(#0,round(ww-dix),round(hh-diy),z);"
k.
elif $3==-1||$3==-2
f ":begin(
const point_x=(($1*-1)*.5+.5)*w;
const point_y=($2*.5+.5)*h;
const inv_point_x=w-point_x;
const inv_point_y=h-point_y;
const cut_ang_s0=abs(atan2(inv_point_y,inv_point_x)*180/pi);
const cut_ang_s1=180-abs(atan2(inv_point_y,point_x)*180/pi);
const cut_ang_s2=180+abs(atan2(point_y,point_x)*180/pi);
const cut_ang_s3=360-abs(atan2(point_y,inv_point_x)*180/pi);
distanceaway(value)=(
value==0?ww-point_x:
value==1?hh-point_y:
value==2?point_x:
point_y;
);
);
surface_angle=(x/w)*360;
surface_angle>cut_ang_s0&&surface_angle<=cut_ang_s1?side=1:
surface_angle>cut_ang_s1&&surface_angle<=cut_ang_s2?side=2:
surface_angle>cut_ang_s2&&surface_angle<=cut_ang_s3?side=3:
side=0;
mdist=abs(side%2?1/sin(surface_angle):1/cos(surface_angle));
dix=(point_x+cos(surface_angle)*distanceaway(side)*mdist*y/h)*((w-1)/w);
diy=(point_y+sin(surface_angle)*distanceaway(side)*mdist*y/h)*((h-1)/h);
I(w-(dix+1),h-(diy+1),z,2);
"
if $3<-1 r2dx 50%,6 fi
elif $3==0||$3==1
if $3>0 r2dx 200%,6 fi
f ":begin(
const ww=w-1;
const hh=h-1;
const sd=max(w,h)/min(w,h);
const sx=w>h?sd:1;
const sy=w>h?1:sd;
const cx=.5+$1*.5;
const cy=.5+$2*.5;
const px=cx*w;
const py=(1-cy)*h;
const sxl=(w/2)/px;
const sxr=(w/2)/(w-px);
const syt=(h/2)/py;
const syb=(h/2)/(h-py);
);
atx=(x/ww-cx)*sx;
aty=(y/hh-(1-cy))*sy;
sur_atan=(atan2(aty,atx)+pi)/(2*pi);
xl=-1+(x/ww)*2*sxl;
xr=1-(1-x/ww)*2*sxr;
yt=-1+(y/hh)*2*syt;
yb=1-(1-y/hh)*2*syb;
xx=x>=px?xr:xl;
yy=y>=py?yb:yt;
sur_max=max(abs(xx),abs(yy));
I(sur_atan*w,sur_max*h,z,2,2);
"
if $3>0 sharpen 2 fi
elif $3==2
$=val
orientation=${val{$>+4}}
perimeter={w}
length_1={$perimeter-h*(4/3)}
length_2={$perimeter-$length_1}
if $orientation
width={min($length_1,$length_2)}
height={max($length_1,$length_2)}
else
width={max($length_1,$length_2)}
height={min($length_1,$length_2)}
fi
v + echo {w},{h} v -
{$width/2},{$height/2},{d},{s},":begin(
const ww=w#0-1;
const hh=h#0-1;
const nw=w-1;
const nh=h-1;
const ox="$width";
const oy="$height";
const sd=max(ox,oy)/min(ox,oy);
if(w>h,
const sxf=ox>oy?sd:1;
const syf=ox>oy?1:sd;
,
const sxf=ox<oy?1:sd;
const syf=ox<oy?sd:1;
);
const cx=.5+$1*.5;
const cy=.5+$2*.5;
const px=cx*(ww-1);
const py=(1-cy)*(hh-1);
const sxl=(ww/2)/px;
const sxr=(ww/2)/(ww-px);
const syt=(hh/2)/py;
const syb=(hh/2)/(hh-py);
);
xx=(x/nw)*(ww);
yy=(y/nh)*(hh);
xl=-1+(xx/ww)*2*sxl;
xr=1-(1-xx/ww)*2*sxr;
yt=-1+(yy/hh)*2*syt;
yb=1-(1-yy/hh)*2*syb;
nxx=xx>px?xr:xl;
nyy=yy>py?yb:yt;
ay=max(abs(nxx),abs(nyy));
ax=(atan2((yy/hh-(1-cy))*syf,(xx/ww-cx)*sxf)+pi)/(2*pi);
I(#0,abs(ax*ww),abs(ay*hh),z,2,1);
"
k.
else error "$3|"$"3!=intnum[-3,2]"
fi
endl done
EDIT: Okay, there seem to be difference now that I analyzed a different way. Now, I’m having the micro-shift problem caused by atan2. I don’t know how to fix that though. 
Here’s an analysis of difference. I decided to try to map coordinate into image, and then find the difference between x,y. This is the result. The red is x, and the y is green. Max difference is 3 pixels off. That’s not good.
1/17/2021:
Looks like I have a much better version of rep_recpoltran now. My next step is to make changes to axis_streak to address the shortcomings with the new rep_recpoltrans, and finally a even better version of Perspective Streak. One that is not that slow and preserve details better.
This is different from what I envisioned. Maybe I am thinking of
