Block 8 has strict standard length for a commercial.
Block 8 score: Before - 1.00 After - 3.00
x3 ??! That is far too much for something so insignificant. I can't find where to change this either.
Generated using Comskip 0.80.031
disable_heuristics=112
global_modifier=1.5
max_commercialbreak=300
min_commercialbreak=55
max_commercial_size=120
min_commercial_size=5 ; 10 does the same thing
min_show_segment_length=120
min_commercial_break_at_start_or_end=55
punish=21
punish_modifier=1.1
global_modifier=1.5
max_schange_modifier=1.5
ar_wrong_modifier=1.5
The real problem here (in case you want to know) is that scene 7 is a (relatively) low brightness ~5 second channel identification scene, and block 8 is a black and white fast action sequence of the show. Very difficult to detect in the 1st place, but it gets worse...
Block 7 has non-strict standard length for a commercial.
Block 7 score: Before - 1.00 After - 1.50
Block 7 is much brighter than average.
Block 7 score: Before - 1.50 After - 1.65
Block 7 is much louder than average.
Block 7 score: Before - 1.65 After - 1.82
Block 7 has more scene change than average.
Block 7 score: Before - 1.82 After - 2.00
...<see above>
Combining Blocks 7 thru 8 result in strict standard commercial length of 19.96 w
ith a tolerance of 1.000000.
Block 8 score: Before - 3.00 After - 4.50 Combined count - 1
Block 7 score: Before - 2.00 After - 2.99 Combined count - 1
Block 8 is much brighter than average.
Block 8 score: Before - 4.50 After - 4.95
Block 8 has more scene change than average.
Block 8 score: Before - 4.95 After - 5.45
Block 8 gets *another* increase just because it can be combined with Block 7. Note that scene 7 would have (correctly) scored higher anyway - it was too loud as well.
This isn't the 1st time irrational score increases from matching some arbitrary length (which dont even appear to have anything to do with the m*_commercial* settings) have really screwed up the otherwise perfect heuristics for me. Are these strict/non-strict length settings hard-coded or...?
(strict) standard lengths are weighted badly
Re: (strict) standard lengths are weighted badly
When you set
verbose=10
the start of the log file lists all tunable parameters with the values used for that comskip run.
You can copy the list of parameters to comskip.ini and change any of them.
Have a look at all the "modifier" parameters.
They are all configurable.
And after having analysed thousands of commercial breaks I tend to disagree with your first statement.
You are in a rare situation when an ontherwise very relevant observation now leads to a wrong conclusion.
verbose=10
the start of the log file lists all tunable parameters with the values used for that comskip run.
You can copy the list of parameters to comskip.ini and change any of them.
Have a look at all the "modifier" parameters.
They are all configurable.
And after having analysed thousands of commercial breaks I tend to disagree with your first statement.
You are in a rare situation when an ontherwise very relevant observation now leads to a wrong conclusion.
Re: (strict) standard lengths are weighted badly
length_strict_modifier=3
length_nonstrict_modifier=1.5
combined_length_strict_modifier=2
combined_length_nonstrict_modifier=1.25
Bingo. But what is the definition of a 'strict' or 'non-strict' length?
length_nonstrict_modifier=1.5
combined_length_strict_modifier=2
combined_length_nonstrict_modifier=1.25
Bingo. But what is the definition of a 'strict' or 'non-strict' length?
Re: (strict) standard lengths are weighted badly
There is a table inside comskip that defines certain durations as strict, others as non strict
Its based on heuristics from hundreds of recordings
Its based on heuristics from hundreds of recordings