| Message ID | 20230816052006.53562-1-leo.izen@gmail.com |
|---|---|
| State | Accepted |
| Commit | 7098bec73bd3de63f514ba6ec091fd2f5096ddce |
| Headers | show |
| Series | [FFmpeg-devel] avcodec/exr: tag gamma=1.0 output as linear light | expand |
| Context | Check | Description |
|---|---|---|
| yinshiyou/make_loongarch64 | success | Make finished |
| yinshiyou/make_fate_loongarch64 | success | Make fate finished |
| andriy/make_x86 | success | Make finished |
| andriy/make_fate_x86 | success | Make fate finished |
LGTM
ons 2023-08-16 klockan 01:20 -0400 skrev Leo Izen: > By default the OpenEXR decoder outputs linear light pixel data by > applying a gamma=1.0 transfer (i.e. a no-op). When it does so, it > should tag the data as linear so color-managed filters or other tools > can work with it correctly. > > Signed-off-by: Leo Izen <leo.izen@gmail.com> > --- > libavcodec/exr.c | 2 ++ > 1 file changed, 2 insertions(+) > > diff --git a/libavcodec/exr.c b/libavcodec/exr.c > index fae1d08ab0..518066facf 100644 > --- a/libavcodec/exr.c > +++ b/libavcodec/exr.c > @@ -2088,6 +2088,8 @@ static int decode_frame(AVCodecContext *avctx, > AVFrame *picture, > > if (s->apply_trc_type != AVCOL_TRC_UNSPECIFIED) > avctx->color_trc = s->apply_trc_type; > + else if (s->gamma > 0.9999f && s->gamma < 1.0001f) > + avctx->color_trc = AVCOL_TRC_LINEAR; I'm going to be difficult here and point out that gamma=0.99991 is not linear. It's probably linear *enough* most of the time, but also 1.0 can be exactly represented by float so an equality check seems appropriate /Tomas
On 8/17/23 08:59, Tomas Härdin wrote: > ons 2023-08-16 klockan 01:20 -0400 skrev Leo Izen: >> By default the OpenEXR decoder outputs linear light pixel data by >> applying a gamma=1.0 transfer (i.e. a no-op). When it does so, it >> should tag the data as linear so color-managed filters or other tools >> can work with it correctly. >> >> Signed-off-by: Leo Izen <leo.izen@gmail.com> >> --- >> libavcodec/exr.c | 2 ++ >> 1 file changed, 2 insertions(+) >> >> diff --git a/libavcodec/exr.c b/libavcodec/exr.c >> index fae1d08ab0..518066facf 100644 >> --- a/libavcodec/exr.c >> +++ b/libavcodec/exr.c >> @@ -2088,6 +2088,8 @@ static int decode_frame(AVCodecContext *avctx, >> AVFrame *picture, >> >> if (s->apply_trc_type != AVCOL_TRC_UNSPECIFIED) >> avctx->color_trc = s->apply_trc_type; >> + else if (s->gamma > 0.9999f && s->gamma < 1.0001f) >> + avctx->color_trc = AVCOL_TRC_LINEAR; > > I'm going to be difficult here and point out that gamma=0.99991 is not > linear. It's probably linear *enough* most of the time, but also 1.0 > can be exactly represented by float so an equality check seems > appropriate. This exact check exists elsewhere in the source file. There's a branch where it sets up a LUT if gamma is not between those values, and has a no-op track otherwise - so in the event you request 0.99995 or something of that form, the code that does the color conversion treats it as 1.0f. - Leo Izen
tor 2023-08-17 klockan 14:04 -0400 skrev Leo Izen: > On 8/17/23 08:59, Tomas Härdin wrote: > > ons 2023-08-16 klockan 01:20 -0400 skrev Leo Izen: > > > By default the OpenEXR decoder outputs linear light pixel data by > > > applying a gamma=1.0 transfer (i.e. a no-op). When it does so, it > > > should tag the data as linear so color-managed filters or other > > > tools > > > can work with it correctly. > > > > > > Signed-off-by: Leo Izen <leo.izen@gmail.com> > > > --- > > > libavcodec/exr.c | 2 ++ > > > 1 file changed, 2 insertions(+) > > > > > > diff --git a/libavcodec/exr.c b/libavcodec/exr.c > > > index fae1d08ab0..518066facf 100644 > > > --- a/libavcodec/exr.c > > > +++ b/libavcodec/exr.c > > > @@ -2088,6 +2088,8 @@ static int decode_frame(AVCodecContext > > > *avctx, > > > AVFrame *picture, > > > > > > if (s->apply_trc_type != AVCOL_TRC_UNSPECIFIED) > > > avctx->color_trc = s->apply_trc_type; > > > + else if (s->gamma > 0.9999f && s->gamma < 1.0001f) > > > + avctx->color_trc = AVCOL_TRC_LINEAR; > > > > I'm going to be difficult here and point out that gamma=0.99991 is > > not > > linear. It's probably linear *enough* most of the time, but also > > 1.0 > > can be exactly represented by float so an equality check seems > > appropriate. > > This exact check exists elsewhere in the source file. There's a > branch > where it sets up a LUT if gamma is not between those values, and has > a > no-op track otherwise - so in the event you request 0.99995 or > something > of that form, the code that does the color conversion treats it as > 1.0f. Alright. I just find myself wondering how gamma could ever become some value that is close to 1.0 but not quite equal. /Tomas
diff --git a/libavcodec/exr.c b/libavcodec/exr.c index fae1d08ab0..518066facf 100644 --- a/libavcodec/exr.c +++ b/libavcodec/exr.c @@ -2088,6 +2088,8 @@ static int decode_frame(AVCodecContext *avctx, AVFrame *picture, if (s->apply_trc_type != AVCOL_TRC_UNSPECIFIED) avctx->color_trc = s->apply_trc_type; + else if (s->gamma > 0.9999f && s->gamma < 1.0001f) + avctx->color_trc = AVCOL_TRC_LINEAR; switch (s->compression) { case EXR_RAW:
By default the OpenEXR decoder outputs linear light pixel data by applying a gamma=1.0 transfer (i.e. a no-op). When it does so, it should tag the data as linear so color-managed filters or other tools can work with it correctly. Signed-off-by: Leo Izen <leo.izen@gmail.com> --- libavcodec/exr.c | 2 ++ 1 file changed, 2 insertions(+)