<lisa79>
Please reply there on github if interested
<lisa79>
The reason is that GME in ffmpeg not supports all VGM formats that are supported by libvgm
<Lynne>
which codecs?
<lisa79>
I mean not all chips that VGM can play. i heard this from libvgm developer
<lisa79>
His answer on vgmrips.net forum to me (private message)
<lisa79>
I quickly skimmed through the GitHub issue you linked:
<lisa79>
ffmpeg may support VGMs through GameMusicEmu. However this will get you only support for uncompressed VGM files and only Master System/MegaDrive hardware is supported.
<lisa79>
The GME version used in ffmpeg and many other open-source products is from about 2007 and lacks support for anything other than SN76489, YM2413 and YM2612.
<lisa79>
Here is also my email: vgm-girl@tutanota.com
<lisa79>
I have to sleep now, it's late
<lisa79>
I hope somebody wants to look at it
<lisa79>
The reason is for having all VGM formats in ffmpeg so ffmpeg can be used in aproject like Mixxx. See github link above. Thanks!
lisa79 has quit [Quit: Connection closed]
kurosu has joined #ffmpeg-devel
wangbin has joined #ffmpeg-devel
SystemError has quit [Read error: Connection reset by peer]
SystemError has joined #ffmpeg-devel
HarshK23 has joined #ffmpeg-devel
AbleBacon_ has quit [Read error: Connection reset by peer]
navi has joined #ffmpeg-devel
philipl has quit [Remote host closed the connection]
philipl has joined #ffmpeg-devel
Teukka has quit [Read error: Connection reset by peer]
Teukka has joined #ffmpeg-devel
Teukka has joined #ffmpeg-devel
Teukka has quit [Changing host]
ccawley2011 has joined #ffmpeg-devel
Krowl has joined #ffmpeg-devel
qeed_ has joined #ffmpeg-devel
qeed has quit [Ping timeout: 255 seconds]
jamrial has joined #ffmpeg-devel
wangbin has quit [Quit: Connection closed]
SystemError has quit [Remote host closed the connection]
SystemError has joined #ffmpeg-devel
mkver has joined #ffmpeg-devel
<courmisch>
Lynne: yeah I ordered one but 512 MiB RAM, ouch
<JEEB>
yea that is not much
ophi has joined #ffmpeg-devel
Krowl has quit [Read error: Connection reset by peer]
<jamrial>
iamf in isobmff defines the entire sequence (up to 18 streams) should be contained in a single trak
<jamrial>
our mp4 demuxer expects a trak to have a single stream
<jamrial>
fun
elastic_dog has quit [Ping timeout: 240 seconds]
<haasn>
query_formats really ought to call into some default function at the end
<haasn>
will it cause problems to just literally call into ff_default_query_formats()?
<haasn>
I suppose we could make this function check to see if it was already set, and if so, skip re-setting it
<haasn>
that way it can work as a base case for all filters that don't care about setting something specific
Krowl has joined #ffmpeg-devel
<JEEB>
Lynne: also this is the current reference sample suite for mpeg-4 audio https://standards.iso.org/iso-iec/14496/-26/ed-2/en/ , looking at the > mpeg4audio-conformance2ndEd_20230125.md5 file name, it's pretty recent. but for USAC specifically the zip for 23003-3 should be enough
cone-261 has joined #ffmpeg-devel
<cone-261>
ffmpeg Niklas Haas master:2d1aeba8f268: avfilter/vf_scale: fix interlaced chroma for other formats
elastic_dog has joined #ffmpeg-devel
<elenril>
mkver: I'm trying to review the rest of the refstruct set, but its size makes that unnecessarily hard
<elenril>
could you split it into smaller pieces?
kekePower has joined #ffmpeg-devel
<durandal_1707>
elenril: its hard to do it
<mkver>
You only need to review the part that was on the ML; the rest will be sent to the ML later.
<mkver>
haasn: I proposed this once, but it was rejected.
<haasn>
mkver: do you have a link to the rejection?
<mkver>
Not atm.
<haasn>
I'm about to add a whole lot more format lists to AVFilterFormatsConfig and I'd rather not go through adding a dozen lines of boilerplate to every filter in existence
<haasn>
and it seems not future proof, anyway
<mkver>
Lots of audio filters that use a query_formats function only set exactly one of things to a different value than ff_default_query_formats() would do, so I proposed that they only set this and the rest gets set generically.
<haasn>
Well, if there were objections to this in the past, I'm sure they will be raised again, and we can discuss it out on the ML
<haasn>
It seems like a design with a bunch of obvious benefits
<durandal_1707>
only NG rejected it
<elenril>
mkver: in your branch it's mixed with the other patches though
<haasn>
what's next, ability to encode HDR into VP8 in nonstrict mode?
<durandal_1707>
limited mjpeg is standard
<haasn>
durandal_1707: then why do we require -strict -1 to encode it?
<durandal_1707>
how other software deals with such files?
<haasn>
other software, like libav?
<durandal_1707>
nope
<durandal_1707>
other mjpeg implementations
<durandal_1707>
do they display limited range mjpeg?
<durandal_1707>
gimp/imagemagick/etc...
<elenril>
mkver: "just" 36 patches, some of which were never sent to the ML
<elenril>
smaller, more digestible sets would make everyone's life easier IMO
<mkver>
I only count 31 patches, two of which have never been sent to the ML.
<elenril>
e.g. you could send everything up to "avcodec/vp9: Join extradata buffer pools"
<elenril>
the once that is approved the rest of your link above
<elenril>
*then
<elenril>
reviewing is hard enough without the added psychological pressure of giant email threads
<courmisch>
Writings of Mass Demotivation
<elenril>
haasn: so why are you not adding color range negotiation to lavfi?
<haasn>
elenril: I am
<elenril>
i thought that was the plan
<elenril>
which patch is that?
<haasn>
the one on my machine
<elenril>
ah, ok then
<jamrial>
black box patch
<elenril>
i was just looking at the RFC from oct 11
<elenril>
heisenpatch
<haasn>
that one is WIP
<haasn>
I cannot for the life of me figure out how to extend AVFilterFormatsConfig in a way that doesn't result in a megapatch
<haasn>
because afaict I need to add handling for the new field to every single place that touches filter configuration in one go, otherwise random stuff will break
<elenril>
The Rampage Of The 9001-line Patch
<durandal_1707>
and how do you plan to make auto-negotiation (very stupid idea when stuff needs to be explicit) for color ranges?
<elenril>
coming soon, to a mailbox near you
<durandal_1707>
this auto-negotiation should be reduced to basic bare-bones stuff
<haasn>
durandal_1707: color ranges is trivial, there are only two cases
<haasn>
the real question is, how do we negotiate between color matrices, is there a "goodness" metric we can use?
<durandal_1707>
and anything else needs user intervention
<haasn>
but the current logic is also trivial
<elenril>
huh, AVFilterFormatsConfig is public?
<haasn>
if there is an intersection, pick the first one in the list that intersects
<elenril>
for what purpose
<mkver>
elenril: It is defined in a public header.
<Lynne>
I did a double take when I read the proposal
<JEEB>
funky
<durandal_1707>
my work is far more important
<another|>
sooo... you can launch CUDA programs from vulkan?
<JEEB>
yea you give it the entry point func pointer essentially
<durandal_1707>
vulkan is tainted binary blob
Krowl has joined #ffmpeg-devel
mkver has joined #ffmpeg-devel
psykose has quit [Remote host closed the connection]
psykose has joined #ffmpeg-devel
cone-261 has quit [Quit: transmission timeout]
Krowl has quit [Read error: Connection reset by peer]
<durandal_1707>
thardin: once i run PCA and get eigenvectors/eigenvalues how to make use of that?
<thardin>
durandal_1707: I'm not 100% sure how apply it to MLP but the first vector captures the most information. if you do the scalar product between it and all samples then quantize that, that will be the best 1-D approximation of the signal
<thardin>
YUV can be seen as a shitty form of PCA, where the Y component captures most variance in typical signals
<thardin>
and mid/side for that matter
<thardin>
strictly speaking PCA gives you a rotation
<durandal_1707>
thardin: but here i allowed only to combine linearly with fixed NxN matrix (N == nb_channels) , all elements of vector of samples are multiplied with same number
<thardin>
uh.. that doesn't sound like NXN
<thardin>
you mean each vector of samples is multiplied by the matrix? because that's what I'd expect
<thardin>
like A*S where S the the matrix formed by all samples in a frame
<thardin>
as in computing A*s for each time instant, where s is the vector of samples in each channel at that instant
<thardin>
but previously you said A is of a particular form
<durandal_1707>
each sample of channel (in block) can be multiplied with only single scalar
<durandal_1707>
and that scalar never changes within this block
<thardin>
so A is diagonal
<thardin>
that doesn't sound very useful
<thardin>
just sounds like per-channel quantization. kind of like mp2 with each subband having its own quantizer
<durandal_1707>
thardin: but you can combine other channels with different scalar, again to produce smaller energy channel
<thardin>
that just sounds like a matrix operation
<durandal_1707>
so matrix transform is Nb_matrix_channels X Nb_matrix_channels
<durandal_1707>
in affine way applied
<thardin>
yeah and if you let the matrix be the outpu of PCA most energy will be comtained in the principal component
<thardin>
that is in the first channel after transforming
<durandal_1707>
but in PCA coefficients can never become negative
<thardin>
yes they can
<thardin>
in fact you typically subtract the mean from the input before you compute it
<durandal_1707>
but for out of phase channels
<durandal_1707>
L = -R
<thardin>
yeah?
<thardin>
you'll get the principal vector [1;-1]
<thardin>
and no residual
<thardin>
what PCA does is try to compress all variance into the first component. then the residual into the second. then the residual of that into the third. and so on
<durandal_1707>
and it can work for stereo: (40*128) X 2 ?
<thardin>
uh yeah?
<durandal_1707>
because its loooks like PCA is pretty slow operation
<thardin>
in general yes
<durandal_1707>
like O(n^3)
<thardin>
for a 2x2 system you can solve it exactly. you jut need to compute the crosscorrelation quickly
elastic_dog has quit [Ping timeout: 245 seconds]
<thardin>
you can do that in two passes I think. maybe even one
<jamrial>
side data being a flat array is so constraining...
AbleBacon_ has joined #ffmpeg-devel
AbleBacon has quit [Ping timeout: 255 seconds]
elastic_dog has joined #ffmpeg-devel
ccawley2011 has quit [Read error: Connection reset by peer]
<durandal_1707>
ffmpeg is dead, 6.1 not comming any time soon
<durandal_1707>
going to make the ultimate fork
navi has quit [Quit: WeeChat 4.0.4]
<thardin>
forking and adding FFBIOS
<thardin>
and FFXML
<durandal_1707>
that is kierank, not me
kasper93 has quit [Ping timeout: 252 seconds]
dellas has quit [Remote host closed the connection]
kasper93 has joined #ffmpeg-devel
cone-669 has joined #ffmpeg-devel
<cone-669>
ffmpeg Paul B Mahol master:393d1ee541b1: avcodec/mlpenc: add 2.1 layout support for truehd
microchip_ is now known as BossTanaka
kasper93 has quit [Ping timeout: 255 seconds]
<durandal_1707>
michaelni: stop blocking 6.1
kasper93 has joined #ffmpeg-devel
<durandal_1707>
Good Night, kids!
derpydoo has quit [Read error: Network is unreachable]
<pross>
G'night grandpa
kurosu has quit [Quit: Connection closed for inactivity]