im@s2 questionboxes detection

ffmpegsource2("C:\Users\Seedmanc\Desktop\[765subs] Ganaha Hibiki - Episode 10.mp4")
ConvertToYV12                   # just to make sure we are in right colorspace no matter what the source is
# boxes sizes
w=round(312*width/(1280*2)) *2          # this way we get independent from source resolution
h=round(120*height/(720*2)) *2          # having boxes' coordinates relative instead of absolute
# coordinates of left box
xb=round(196*width/(1280*2)) *2         # also this ensures mod2 rule, since video in YV12 format
yb=round(322*height/(720*2)) *2         # must have even dimensions
# coordinates of right box
xr=round(770*width/(1280*2)) *2
yr=round(322*height/(720*2)) *2
# we don't need third box, as it is not always present
src1= crop(xb,yb,w,h).blur(1.58)        #source video area of left box
bluemask=src1.maskhs(starthue=360-45,endhue=360-7,maxsat=70,minsat=5).converttoyv12.mt_inflate.mt_expand
# ^ selection by color blue
src2= crop(xr,yr,w,h).blur(1.58)        #source video area of right box
redmask = src2. maskhs(starthue=60, endhue=120, maxsat=75, minsat=10 ) .converttoyv12. mt_inflate. mt_expand
# ^ selection by color red
comb=mt_logic(bluemask, redmask, mode="min")
# ^ combining masks by AND operator - so that detection happens only when both boxes are found
scriptclip("""AverageLuma(comb )>245 ? \
 invert : \
 last""")
__END__
Here it simply switches to inverted colors when it detects question boxes appearing, just to show that it works.
245 is the threshold of detection, lower values will make it happen earlier and last longer (since those boxes slide in and out gradually) with the chance of false detection, higher values will do the opposite.