video: add vf_fingerprint and a skip-logo script
skip-logo.lua is just what I wanted to have. Explanations are on the top
of that file. As usual, all documentation threatens to remove this stuff
all the time, since this stuff is just for me, and unlike a normal user
I can afford the luxuary of hacking the shit directly into the player.
vf_fingerprint is needed to support this script. It needs to scale down
video frames as part of its operation. For that, it uses zimg. zimg is
much faster than libswscale and generates more correct output. (The
filter includes a runtime fallback, but it doesn't even work because
libswscale fucks up and can't do YUV->Gray with range adjustment.)
Note on the algorithm: seems almost too simple, but was suggested to me.
It seems to be pretty effective, although long time experience with
false positives is missing. At first I wanted to use dHash [1][2], which
is also pretty simple and effective, but might actually be worse than
the implemented mechanism. dHash has the advantage that the fingerprint
is smaller. But exact matching is too unreliable, and you'd still need
to determine the number of different bits for fuzzier comparison. So
there wasn't really a reason to use it.
[1] https://pypi.org/project/dhash/
[2] http://www.hackerfactor.com/blog/index.php?/archives/529-Kind-of-Like-That.html
2019-07-15 01:20:40 +00:00
|
|
|
--[[
|
|
|
|
|
|
|
|
Automatically skip in files if video frames with pre-supplied fingerprints are
|
|
|
|
detected. This will skip ahead by a pre-configured amount of time if a matching
|
|
|
|
video frame is detected.
|
|
|
|
|
|
|
|
This requires the vf_fingerprint video filter to be compiled in. Read the
|
|
|
|
documentation of this filter for caveats (which will automatically apply to
|
|
|
|
this script as well), such as no support for zero-copy hardware decoding.
|
|
|
|
|
|
|
|
You need to manually gather and provide fingerprints for video frames and add
|
|
|
|
them to a configuration file in script-opts/skip-logo.conf (the "script-opts"
|
|
|
|
directory must be in the mpv configuration directory, typically ~/.config/mpv/).
|
|
|
|
|
|
|
|
Example script-opts/skip-logo.conf:
|
|
|
|
|
|
|
|
|
|
|
|
cases = {
|
|
|
|
{
|
|
|
|
-- Skip ahead 10 seconds if a black frame was detected
|
|
|
|
-- Note: this is dangerous non-sense. It's just for demonstration.
|
|
|
|
name = "black frame", -- print if matched
|
|
|
|
skip = 10, -- number of seconds to skip forward
|
|
|
|
score = 0.3, -- required score
|
|
|
|
fingerprint = "00000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000",
|
|
|
|
},
|
|
|
|
{
|
|
|
|
-- Skip ahead 20 seconds if a white frame was detected
|
|
|
|
-- Note: this is dangerous non-sense. It's just for demonstration.
|
|
|
|
name = "fun2",
|
|
|
|
skip = 20,
|
|
|
|
fingerprint = "ffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffff",
|
|
|
|
},
|
|
|
|
}
|
|
|
|
|
|
|
|
This is actually a lua file. Lua was chosen because it seemed less of a pain to
|
|
|
|
parse. Future versions of this script may change the format.
|
|
|
|
|
|
|
|
The fingerprint is a video frame, converted to "gray" (8 bit per pixels), full
|
|
|
|
range, each pixel concatenated into an array, converted to a hex string. You
|
|
|
|
can produce these fingerprints by running this manually:
|
|
|
|
|
|
|
|
mpv --vf=fingerprint:print yourfile.mkv
|
|
|
|
|
|
|
|
This will log the fingerprint of each video frame to the console, along with its
|
|
|
|
timestamp. You find the fingerprint of a unique-enough looking frame, and add
|
|
|
|
it as entry to skip-logo.conf.
|
|
|
|
|
|
|
|
You can provide a score for "fuzziness". If no score is provided, a default
|
|
|
|
value of 0.3 is used. The score is inverse: 0 means exactly the same, while a
|
|
|
|
higher score means a higher difference. Currently, the score is computed as
|
|
|
|
euclidean distance between the video frame and the pre-provided fingerprint,
|
|
|
|
thus the highest score is 16. You probably want a score lower than 1 at least.
|
|
|
|
(This algorithm is very primitive, but also simple and fast to compute.)
|
|
|
|
|
|
|
|
There's always the danger of false positives, which might be quite annoying.
|
|
|
|
It's up to you what you hate more, the logo, or random skips if false positives
|
|
|
|
are detected. Also, it's always active, and might eat too much CPU with files
|
|
|
|
that have a high resolution or framerate. To temporarily disable the script,
|
|
|
|
having a keybind like this in your input.conf will be helpful:
|
|
|
|
|
|
|
|
ctrl+k vf toggle @skip-logo
|
|
|
|
|
|
|
|
This will disable/enable the fingerprint filter, which the script automatically
|
|
|
|
adds at start.
|
|
|
|
|
|
|
|
Another important caveat is that the script currently disables matching during
|
|
|
|
seeking or playback initialization, which means it cannot match the first few
|
|
|
|
frames of a video. This could be fixed, but the author was too lazy to do so.
|
|
|
|
|
|
|
|
--]]
|
|
|
|
|
|
|
|
local utils = require "mp.utils"
|
|
|
|
local msg = require "mp.msg"
|
|
|
|
|
|
|
|
local label = "skip-logo"
|
|
|
|
local meta_property = string.format("vf-metadata/%s", label)
|
|
|
|
|
|
|
|
local config = {}
|
|
|
|
local cases = {}
|
|
|
|
local cur_bmp
|
skip-logo.lua: fix skipping in the first two frames
mpv typically decodes and filters at least 2 frames before starting
playback. This happens during seeks, as well as when playback starts
from the beginning of the file.
skip-logo.lua receives notifications for all filtered frames, even
during seeking. It should interrupt during seeking, so as a crude
heuristic, it ignored all frames while the player was seeking. This does
not mean all these frames are skipped due to seeking (thus it's a "crude
hueristic"). In particular, it means that the first 2 frames of a video
cannot be skipped, since they're filtered within the playback restart
phase (equivalent to "seeking").
Fix this by making the heuristic slightly less crude. Since we observe
the property as "none", the property is not actually read until we do it
explicitly. By not reading it during seeking, we can let the frames
internally queue up (vf_fingerprint discards them in a ringbuffer-like
fashion if they're too many). Then, if seeking ends, we get the current
playback timestamp, and check queued up frames that are at or after that
timestamp. (In some ways, this duplicates what the player's seeking
logic does.)
A disadvantage is that this is racy. While playback-time is guaranteed
to be set when seeking changes from false to true, playback could
already have progressed to the next frame (or more) before the script
gets time to react. In theory, we could add a seek restart hook or so,
but I don't want to. A property that returns the last playback restart
time would also do it, but feels to special. Not an important problem
in practice anyway.
2019-10-08 19:26:17 +00:00
|
|
|
local seeking = false
|
|
|
|
local playback_start_pts = nil
|
video: add vf_fingerprint and a skip-logo script
skip-logo.lua is just what I wanted to have. Explanations are on the top
of that file. As usual, all documentation threatens to remove this stuff
all the time, since this stuff is just for me, and unlike a normal user
I can afford the luxuary of hacking the shit directly into the player.
vf_fingerprint is needed to support this script. It needs to scale down
video frames as part of its operation. For that, it uses zimg. zimg is
much faster than libswscale and generates more correct output. (The
filter includes a runtime fallback, but it doesn't even work because
libswscale fucks up and can't do YUV->Gray with range adjustment.)
Note on the algorithm: seems almost too simple, but was suggested to me.
It seems to be pretty effective, although long time experience with
false positives is missing. At first I wanted to use dHash [1][2], which
is also pretty simple and effective, but might actually be worse than
the implemented mechanism. dHash has the advantage that the fingerprint
is smaller. But exact matching is too unreliable, and you'd still need
to determine the number of different bits for fuzzier comparison. So
there wasn't really a reason to use it.
[1] https://pypi.org/project/dhash/
[2] http://www.hackerfactor.com/blog/index.php?/archives/529-Kind-of-Like-That.html
2019-07-15 01:20:40 +00:00
|
|
|
|
|
|
|
-- Convert a hex string to an array. Convert each byte to a [0,1] float by
|
|
|
|
-- interpreting it as normalized uint8_t.
|
|
|
|
-- The data parameter, if not nil, may be used as storage (avoiding garbage).
|
|
|
|
local function hex_to_norm8(hex, data)
|
|
|
|
local size = math.floor(#hex / 2)
|
|
|
|
if #hex ~= size * 2 then
|
|
|
|
return nil
|
|
|
|
end
|
|
|
|
local res
|
|
|
|
if (data ~= nil) and (#data == size) then
|
|
|
|
res = data
|
|
|
|
else
|
|
|
|
res = {}
|
|
|
|
end
|
|
|
|
for i = 1, size do
|
|
|
|
local num = tonumber(hex:sub(i * 2, i * 2 + 1), 16)
|
|
|
|
if num == nil then
|
|
|
|
return nil
|
|
|
|
end
|
|
|
|
res[i] = num / 255.0
|
|
|
|
end
|
|
|
|
return res
|
|
|
|
end
|
|
|
|
|
|
|
|
local function compare_bmp(a, b)
|
|
|
|
if #a ~= #b then
|
|
|
|
return nil -- can't compare
|
|
|
|
end
|
|
|
|
local sum = 0
|
|
|
|
for i = 1, #a do
|
|
|
|
local diff = a[i] - b[i]
|
|
|
|
sum = sum + diff * diff
|
|
|
|
end
|
|
|
|
return math.sqrt(sum)
|
|
|
|
end
|
|
|
|
|
|
|
|
local function load_config()
|
|
|
|
local conf_file = mp.find_config_file("script-opts/skip-logo.conf")
|
|
|
|
local conf_fn
|
|
|
|
local err = nil
|
|
|
|
if conf_file then
|
|
|
|
if setfenv then
|
|
|
|
conf_fn, err = loadfile(conf_file)
|
|
|
|
if conf_fn then
|
|
|
|
setfenv(conf_fn, config)
|
|
|
|
end
|
|
|
|
else
|
|
|
|
conf_fn, err = loadfile(conf_file, "t", config)
|
|
|
|
end
|
|
|
|
else
|
|
|
|
err = "config file not found"
|
|
|
|
end
|
|
|
|
|
|
|
|
if conf_fn and (not err) then
|
|
|
|
local ok, err2 = pcall(conf_fn)
|
|
|
|
err = err2
|
|
|
|
end
|
|
|
|
|
|
|
|
if err then
|
|
|
|
msg.error("Failed to load config file:", err)
|
|
|
|
end
|
|
|
|
|
|
|
|
if config.cases then
|
|
|
|
for n, case in ipairs(config.cases) do
|
|
|
|
local err = nil
|
|
|
|
case.bitmap = hex_to_norm8(case.fingerprint)
|
|
|
|
if case.bitmap == nil then
|
|
|
|
err = "invalid or missing fingerprint field"
|
|
|
|
end
|
|
|
|
if case.score == nil then
|
|
|
|
case.score = 0.3
|
|
|
|
end
|
|
|
|
if type(case.score) ~= "number" then
|
|
|
|
err = "score field is not a number"
|
|
|
|
end
|
|
|
|
if type(case.skip) ~= "number" then
|
|
|
|
err = "skip field is not a number or missing"
|
|
|
|
end
|
|
|
|
if case.name == nil then
|
|
|
|
case.name = ("Entry %d"):format(n)
|
|
|
|
end
|
|
|
|
if err == nil then
|
|
|
|
cases[#cases + 1] = case
|
|
|
|
else
|
|
|
|
msg.error(("Entry %s: %s, ignoring."):format(case.name, err))
|
|
|
|
end
|
|
|
|
end
|
|
|
|
end
|
|
|
|
end
|
|
|
|
|
|
|
|
load_config()
|
|
|
|
|
|
|
|
-- Returns true on match and if something was done.
|
|
|
|
local function check_fingerprint(hex, pts)
|
|
|
|
local bmp = hex_to_norm8(hex, cur_bmp)
|
|
|
|
cur_bmp = bmp
|
|
|
|
|
|
|
|
-- If parsing the filter's result failed (well, it shouldn't).
|
|
|
|
assert(bmp ~= nil, "filter returned nonsense")
|
|
|
|
|
|
|
|
for _, case in ipairs(cases) do
|
|
|
|
local score = compare_bmp(case.bitmap, bmp)
|
|
|
|
if (score ~= nil) and (score <= case.score) then
|
|
|
|
msg.warn(("Matching %s: score=%f (required: %f) at %s, skipping %f seconds"):
|
|
|
|
format(case.name, score, case.score, mp.format_time(pts), case.skip))
|
|
|
|
mp.commandv("seek", pts + case.skip, "absolute+exact")
|
|
|
|
return true
|
|
|
|
end
|
|
|
|
end
|
|
|
|
|
|
|
|
return false
|
|
|
|
end
|
|
|
|
|
skip-logo.lua: fix skipping in the first two frames
mpv typically decodes and filters at least 2 frames before starting
playback. This happens during seeks, as well as when playback starts
from the beginning of the file.
skip-logo.lua receives notifications for all filtered frames, even
during seeking. It should interrupt during seeking, so as a crude
heuristic, it ignored all frames while the player was seeking. This does
not mean all these frames are skipped due to seeking (thus it's a "crude
hueristic"). In particular, it means that the first 2 frames of a video
cannot be skipped, since they're filtered within the playback restart
phase (equivalent to "seeking").
Fix this by making the heuristic slightly less crude. Since we observe
the property as "none", the property is not actually read until we do it
explicitly. By not reading it during seeking, we can let the frames
internally queue up (vf_fingerprint discards them in a ringbuffer-like
fashion if they're too many). Then, if seeking ends, we get the current
playback timestamp, and check queued up frames that are at or after that
timestamp. (In some ways, this duplicates what the player's seeking
logic does.)
A disadvantage is that this is racy. While playback-time is guaranteed
to be set when seeking changes from false to true, playback could
already have progressed to the next frame (or more) before the script
gets time to react. In theory, we could add a seek restart hook or so,
but I don't want to. A property that returns the last playback restart
time would also do it, but feels to special. Not an important problem
in practice anyway.
2019-10-08 19:26:17 +00:00
|
|
|
local function read_frames()
|
video: add vf_fingerprint and a skip-logo script
skip-logo.lua is just what I wanted to have. Explanations are on the top
of that file. As usual, all documentation threatens to remove this stuff
all the time, since this stuff is just for me, and unlike a normal user
I can afford the luxuary of hacking the shit directly into the player.
vf_fingerprint is needed to support this script. It needs to scale down
video frames as part of its operation. For that, it uses zimg. zimg is
much faster than libswscale and generates more correct output. (The
filter includes a runtime fallback, but it doesn't even work because
libswscale fucks up and can't do YUV->Gray with range adjustment.)
Note on the algorithm: seems almost too simple, but was suggested to me.
It seems to be pretty effective, although long time experience with
false positives is missing. At first I wanted to use dHash [1][2], which
is also pretty simple and effective, but might actually be worse than
the implemented mechanism. dHash has the advantage that the fingerprint
is smaller. But exact matching is too unreliable, and you'd still need
to determine the number of different bits for fuzzier comparison. So
there wasn't really a reason to use it.
[1] https://pypi.org/project/dhash/
[2] http://www.hackerfactor.com/blog/index.php?/archives/529-Kind-of-Like-That.html
2019-07-15 01:20:40 +00:00
|
|
|
local result = mp.get_property_native(meta_property)
|
|
|
|
if result == nil then
|
|
|
|
return
|
|
|
|
end
|
|
|
|
|
|
|
|
-- Try to get all entries. Out of laziness, assume that there are at most
|
|
|
|
-- 100 entries. (In fact, vf_fingerprint limits it to 10.)
|
|
|
|
for i = 0, 99 do
|
|
|
|
local prefix = string.format("fp%d.", i)
|
|
|
|
local hex = result[prefix .. "hex"]
|
|
|
|
|
|
|
|
local pts = tonumber(result[prefix .. "pts"])
|
|
|
|
if (hex == nil) or (pts == nil) then
|
|
|
|
break
|
|
|
|
end
|
|
|
|
|
skip-logo.lua: fix skipping in the first two frames
mpv typically decodes and filters at least 2 frames before starting
playback. This happens during seeks, as well as when playback starts
from the beginning of the file.
skip-logo.lua receives notifications for all filtered frames, even
during seeking. It should interrupt during seeking, so as a crude
heuristic, it ignored all frames while the player was seeking. This does
not mean all these frames are skipped due to seeking (thus it's a "crude
hueristic"). In particular, it means that the first 2 frames of a video
cannot be skipped, since they're filtered within the playback restart
phase (equivalent to "seeking").
Fix this by making the heuristic slightly less crude. Since we observe
the property as "none", the property is not actually read until we do it
explicitly. By not reading it during seeking, we can let the frames
internally queue up (vf_fingerprint discards them in a ringbuffer-like
fashion if they're too many). Then, if seeking ends, we get the current
playback timestamp, and check queued up frames that are at or after that
timestamp. (In some ways, this duplicates what the player's seeking
logic does.)
A disadvantage is that this is racy. While playback-time is guaranteed
to be set when seeking changes from false to true, playback could
already have progressed to the next frame (or more) before the script
gets time to react. In theory, we could add a seek restart hook or so,
but I don't want to. A property that returns the last playback restart
time would also do it, but feels to special. Not an important problem
in practice anyway.
2019-10-08 19:26:17 +00:00
|
|
|
local skip = false -- blame Lua for not having "continue" or "goto", not me
|
|
|
|
|
|
|
|
-- If seeking just stopped, there will be frames before the seek target,
|
|
|
|
-- ignore them by checking the timestamps.
|
|
|
|
if playback_start_pts ~= nil then
|
|
|
|
if pts >= playback_start_pts then
|
|
|
|
playback_start_pts = nil -- just for robustness
|
|
|
|
else
|
|
|
|
skip = true
|
|
|
|
end
|
|
|
|
end
|
|
|
|
|
|
|
|
if not skip then
|
|
|
|
if check_fingerprint(hex, pts) then
|
|
|
|
break
|
|
|
|
end
|
video: add vf_fingerprint and a skip-logo script
skip-logo.lua is just what I wanted to have. Explanations are on the top
of that file. As usual, all documentation threatens to remove this stuff
all the time, since this stuff is just for me, and unlike a normal user
I can afford the luxuary of hacking the shit directly into the player.
vf_fingerprint is needed to support this script. It needs to scale down
video frames as part of its operation. For that, it uses zimg. zimg is
much faster than libswscale and generates more correct output. (The
filter includes a runtime fallback, but it doesn't even work because
libswscale fucks up and can't do YUV->Gray with range adjustment.)
Note on the algorithm: seems almost too simple, but was suggested to me.
It seems to be pretty effective, although long time experience with
false positives is missing. At first I wanted to use dHash [1][2], which
is also pretty simple and effective, but might actually be worse than
the implemented mechanism. dHash has the advantage that the fingerprint
is smaller. But exact matching is too unreliable, and you'd still need
to determine the number of different bits for fuzzier comparison. So
there wasn't really a reason to use it.
[1] https://pypi.org/project/dhash/
[2] http://www.hackerfactor.com/blog/index.php?/archives/529-Kind-of-Like-That.html
2019-07-15 01:20:40 +00:00
|
|
|
end
|
|
|
|
end
|
skip-logo.lua: fix skipping in the first two frames
mpv typically decodes and filters at least 2 frames before starting
playback. This happens during seeks, as well as when playback starts
from the beginning of the file.
skip-logo.lua receives notifications for all filtered frames, even
during seeking. It should interrupt during seeking, so as a crude
heuristic, it ignored all frames while the player was seeking. This does
not mean all these frames are skipped due to seeking (thus it's a "crude
hueristic"). In particular, it means that the first 2 frames of a video
cannot be skipped, since they're filtered within the playback restart
phase (equivalent to "seeking").
Fix this by making the heuristic slightly less crude. Since we observe
the property as "none", the property is not actually read until we do it
explicitly. By not reading it during seeking, we can let the frames
internally queue up (vf_fingerprint discards them in a ringbuffer-like
fashion if they're too many). Then, if seeking ends, we get the current
playback timestamp, and check queued up frames that are at or after that
timestamp. (In some ways, this duplicates what the player's seeking
logic does.)
A disadvantage is that this is racy. While playback-time is guaranteed
to be set when seeking changes from false to true, playback could
already have progressed to the next frame (or more) before the script
gets time to react. In theory, we could add a seek restart hook or so,
but I don't want to. A property that returns the last playback restart
time would also do it, but feels to special. Not an important problem
in practice anyway.
2019-10-08 19:26:17 +00:00
|
|
|
end
|
|
|
|
|
|
|
|
mp.observe_property(meta_property, "none", function()
|
|
|
|
-- Ignore frames that are decoded/filtered during seeking.
|
|
|
|
if seeking then
|
|
|
|
return
|
|
|
|
end
|
|
|
|
|
|
|
|
read_frames()
|
|
|
|
end)
|
|
|
|
|
|
|
|
mp.observe_property("seeking", "bool", function(name, val)
|
|
|
|
seeking = val
|
|
|
|
if seeking == false then
|
|
|
|
playback_start_pts = mp.get_property_number("playback-time")
|
|
|
|
read_frames()
|
|
|
|
end
|
video: add vf_fingerprint and a skip-logo script
skip-logo.lua is just what I wanted to have. Explanations are on the top
of that file. As usual, all documentation threatens to remove this stuff
all the time, since this stuff is just for me, and unlike a normal user
I can afford the luxuary of hacking the shit directly into the player.
vf_fingerprint is needed to support this script. It needs to scale down
video frames as part of its operation. For that, it uses zimg. zimg is
much faster than libswscale and generates more correct output. (The
filter includes a runtime fallback, but it doesn't even work because
libswscale fucks up and can't do YUV->Gray with range adjustment.)
Note on the algorithm: seems almost too simple, but was suggested to me.
It seems to be pretty effective, although long time experience with
false positives is missing. At first I wanted to use dHash [1][2], which
is also pretty simple and effective, but might actually be worse than
the implemented mechanism. dHash has the advantage that the fingerprint
is smaller. But exact matching is too unreliable, and you'd still need
to determine the number of different bits for fuzzier comparison. So
there wasn't really a reason to use it.
[1] https://pypi.org/project/dhash/
[2] http://www.hackerfactor.com/blog/index.php?/archives/529-Kind-of-Like-That.html
2019-07-15 01:20:40 +00:00
|
|
|
end)
|
|
|
|
|
|
|
|
local filters = mp.get_property_native("option-info/vf/choices", {})
|
|
|
|
local found = false
|
|
|
|
for _, f in ipairs(filters) do
|
|
|
|
if f == "fingerprint" then
|
|
|
|
found = true
|
|
|
|
break
|
|
|
|
end
|
|
|
|
end
|
|
|
|
|
|
|
|
if found then
|
|
|
|
mp.command(("no-osd vf add @%s:fingerprint"):format(label, filter))
|
|
|
|
else
|
|
|
|
msg.warn("vf_fingerprint not found")
|
|
|
|
end
|