[FFmpeg-devel] [PATCH] vf_colorspace: Add an option to clamp trc LUT output

[FFmpeg-devel] [PATCH] vf_colorspace: Add an option to clamp trc LUT output

[Drew Dunne via ffmpeg-devel]

Add a new flag to the vf_colorspace filter which provides the user an
option to clamp the linear and delinear transfer characteristics LUT
values to the [0, 1] represented range. This helps constrain the
potential value range when converting between colorspaces.

Signed-off-by: Drew Dunne <asdunne@google.com>
---
 doc/filters.texi            |  3 +++
 libavfilter/vf_colorspace.c | 14 ++++++++++++--
 2 files changed, 15 insertions(+), 2 deletions(-)

diff --git a/doc/filters.texi b/doc/filters.texi
index 908c98a3cf..cb09d73f62 100644
--- a/doc/filters.texi
+++ b/doc/filters.texi
@@ -10403,6 +10403,9 @@ von Kries whitepoint adaptation
 identity whitepoint adaptation (i.e. no whitepoint adaptation)
 @end table
 
+@item clamptrc
+Clamps the linear and delinear transfer characteristics LUT values to the [0, 1] represented range.
+
 @item iall
 Override all input properties at once. Same accepted values as @ref{all}.
 
diff --git a/libavfilter/vf_colorspace.c b/libavfilter/vf_colorspace.c
index e1f4725f63..15b4858b24 100644
--- a/libavfilter/vf_colorspace.c
+++ b/libavfilter/vf_colorspace.c
@@ -123,6 +123,7 @@ typedef struct ColorSpaceContext {
     int fast_mode;
     enum DitherMode dither;
     enum WhitepointAdaptation wp_adapt;
+    int clamp_trc;
 
     int16_t *rgb[3];
     ptrdiff_t rgb_stride;
@@ -215,7 +216,9 @@ static int fill_gamma_table(ColorSpaceContext *s)
         } else {
             d = out_alpha * pow(v, out_gamma) - (out_alpha - 1.0);
         }
-        s->delin_lut[n] = av_clip_int16(lrint(d * 28672.0));
+        long d_rounded = lrint(d * 28672.0);
+        s->delin_lut[n] = s->clamp_trc ? av_clip64(d_rounded, 0, 28672)
+                                       : av_clip_int16(d_rounded); 
 
         // linearize
         if (v <= -in_beta * in_delta) {
@@ -225,7 +228,9 @@ static int fill_gamma_table(ColorSpaceContext *s)
         } else {
             l = pow((v + in_alpha - 1.0) * in_ialpha, in_igamma);
         }
-        s->lin_lut[n] = av_clip_int16(lrint(l * 28672.0));
+        long l_rounded = lrint(l * 28672.0);
+        s->lin_lut[n] = s->clamp_trc ? av_clip64(l_rounded, 0, 28672)
+                                     : av_clip_int16(l_rounded); 
     }
 
     return 0;
@@ -1000,6 +1005,11 @@ static const AVOption colorspace_options[] = {
     ENUM("vonkries", WP_ADAPT_VON_KRIES, "wpadapt"),
     ENUM("identity", WP_ADAPT_IDENTITY, "wpadapt"),
 
+    { "clamptrc", 
+      "Clamps the linear and delinear LUT output values to the range [0, 1].",
+      OFFSET(clamp_trc), AV_OPT_TYPE_BOOL,  { .i64 = 0    },
+      0, 1, FLAGS },
+
     { "iall",       "Set all input color properties together",
       OFFSET(user_iall),   AV_OPT_TYPE_INT, { .i64 = CS_UNSPECIFIED },
       CS_UNSPECIFIED, CS_NB - 1, FLAGS, .unit = "all" },
-- 
2.51.0.rc1.163.g2494970778-goog

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

Re: [FFmpeg-devel] [PATCH] vf_colorspace: Add an option to clamp trc LUT output

[Ronald S. Bultje via ffmpeg-devel]

Hi Drew,

On Mon, Aug 18, 2025 at 11:55 AM Drew Dunne via ffmpeg-devel <
ffmpeg-devel@ffmpeg.org> wrote:

> Add a new flag to the vf_colorspace filter which provides the user an
> option to clamp the linear and delinear transfer characteristics LUT
> values to the [0, 1] represented range. This helps constrain the
> potential value range when converting between colorspaces.
>
> Signed-off-by: Drew Dunne <asdunne@google.com>
> ---
>  doc/filters.texi            |  3 +++
>  libavfilter/vf_colorspace.c | 14 ++++++++++++--
>  2 files changed, 15 insertions(+), 2 deletions(-)
>
> diff --git a/doc/filters.texi b/doc/filters.texi
> index 908c98a3cf..cb09d73f62 100644
> --- a/doc/filters.texi
> +++ b/doc/filters.texi
> @@ -10403,6 +10403,9 @@ von Kries whitepoint adaptation
>  identity whitepoint adaptation (i.e. no whitepoint adaptation)
>  @end table
>
> +@item clamptrc
> +Clamps the linear and delinear transfer characteristics LUT values to the
> [0, 1] represented range.
> +
>  @item iall
>  Override all input properties at once. Same accepted values as @ref{all}.
>
> diff --git a/libavfilter/vf_colorspace.c b/libavfilter/vf_colorspace.c
> index e1f4725f63..15b4858b24 100644
> --- a/libavfilter/vf_colorspace.c
> +++ b/libavfilter/vf_colorspace.c
> @@ -123,6 +123,7 @@ typedef struct ColorSpaceContext {
>      int fast_mode;
>      enum DitherMode dither;
>      enum WhitepointAdaptation wp_adapt;
> +    int clamp_trc;
>
>      int16_t *rgb[3];
>      ptrdiff_t rgb_stride;
> @@ -215,7 +216,9 @@ static int fill_gamma_table(ColorSpaceContext *s)
>          } else {
>              d = out_alpha * pow(v, out_gamma) - (out_alpha - 1.0);
>          }
> -        s->delin_lut[n] = av_clip_int16(lrint(d * 28672.0));
> +        long d_rounded = lrint(d * 28672.0);
> +        s->delin_lut[n] = s->clamp_trc ? av_clip64(d_rounded, 0, 28672)
> +                                       : av_clip_int16(d_rounded);
>
>          // linearize
>          if (v <= -in_beta * in_delta) {
> @@ -225,7 +228,9 @@ static int fill_gamma_table(ColorSpaceContext *s)
>          } else {
>              l = pow((v + in_alpha - 1.0) * in_ialpha, in_igamma);
>          }
> -        s->lin_lut[n] = av_clip_int16(lrint(l * 28672.0));
> +        long l_rounded = lrint(l * 28672.0);
> +        s->lin_lut[n] = s->clamp_trc ? av_clip64(l_rounded, 0, 28672)
> +                                     : av_clip_int16(l_rounded);
>      }
>
>      return 0;
> @@ -1000,6 +1005,11 @@ static const AVOption colorspace_options[] = {
>      ENUM("vonkries", WP_ADAPT_VON_KRIES, "wpadapt"),
>      ENUM("identity", WP_ADAPT_IDENTITY, "wpadapt"),
>
> +    { "clamptrc",
> +      "Clamps the linear and delinear LUT output values to the range [0,
> 1].",
> +      OFFSET(clamp_trc), AV_OPT_TYPE_BOOL,  { .i64 = 0    },
> +      0, 1, FLAGS },
> +
>      { "iall",       "Set all input color properties together",
>        OFFSET(user_iall),   AV_OPT_TYPE_INT, { .i64 = CS_UNSPECIFIED },
>        CS_UNSPECIFIED, CS_NB - 1, FLAGS, .unit = "all" },
> --
> 2.51.0.rc1.163.g2494970778-goog


Can you provide an example where this is relevant? I'm trying to understand
the three components here: changing from int to long, clipping in 64bit
instead of 16bit, and adding positive+negative range clamp. I'm wondering
which of these you've observed, and what type of minimal fix would be
sufficient.

(Part of the question here is that we might not want this to be an
AVOption, but to be always-on.)

Thanks,
Ronald
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

[FFmpeg-devel] [PATCH] vf_colorspace: Add an option to clamp trc LUT output

[Drew Dunne via ffmpeg-devel]

Add a new flag to the vf_colorspace filter which provides the user an
option to clamp the linear and delinear transfer characteristics LUT
values to the [0, 1] represented range. This helps constrain the
potential value range when converting between colorspaces.

Signed-off-by: Drew Dunne <asdunne@google.com>
---
 doc/filters.texi            |  3 +++
 libavfilter/vf_colorspace.c | 14 ++++++++++++--
 2 files changed, 15 insertions(+), 2 deletions(-)

diff --git a/doc/filters.texi b/doc/filters.texi
index 908c98a3cf..cb09d73f62 100644
--- a/doc/filters.texi
+++ b/doc/filters.texi
@@ -10403,6 +10403,9 @@ von Kries whitepoint adaptation
 identity whitepoint adaptation (i.e. no whitepoint adaptation)
 @end table
 
+@item clamptrc
+Clamps the linear and delinear transfer characteristics LUT values to the [0, 1] represented range.
+
 @item iall
 Override all input properties at once. Same accepted values as @ref{all}.
 
diff --git a/libavfilter/vf_colorspace.c b/libavfilter/vf_colorspace.c
index e1f4725f63..ad8596918f 100644
--- a/libavfilter/vf_colorspace.c
+++ b/libavfilter/vf_colorspace.c
@@ -123,6 +123,7 @@ typedef struct ColorSpaceContext {
     int fast_mode;
     enum DitherMode dither;
     enum WhitepointAdaptation wp_adapt;
+    int clamp_trc;
 
     int16_t *rgb[3];
     ptrdiff_t rgb_stride;
@@ -215,7 +216,9 @@ static int fill_gamma_table(ColorSpaceContext *s)
         } else {
             d = out_alpha * pow(v, out_gamma) - (out_alpha - 1.0);
         }
-        s->delin_lut[n] = av_clip_int16(lrint(d * 28672.0));
+        int d_rounded = lrint(d * 28672.0);
+        s->delin_lut[n] = s->clamp_trc ? av_clip(d_rounded, 0, 28672)
+                                       : av_clip_int16(d_rounded);
 
         // linearize
         if (v <= -in_beta * in_delta) {
@@ -225,7 +228,9 @@ static int fill_gamma_table(ColorSpaceContext *s)
         } else {
             l = pow((v + in_alpha - 1.0) * in_ialpha, in_igamma);
         }
-        s->lin_lut[n] = av_clip_int16(lrint(l * 28672.0));
+        int l_rounded = lrint(l * 28672.0);
+        s->lin_lut[n] = s->clamp_trc ? av_clip(l_rounded, 0, 28672)
+                                     : av_clip_int16(l_rounded);
     }
 
     return 0;
@@ -1000,6 +1005,11 @@ static const AVOption colorspace_options[] = {
     ENUM("vonkries", WP_ADAPT_VON_KRIES, "wpadapt"),
     ENUM("identity", WP_ADAPT_IDENTITY, "wpadapt"),
 
+    { "clamptrc",
+      "Clamps the linear and delinear LUT output values to the range [0, 1].",
+      OFFSET(clamp_trc), AV_OPT_TYPE_BOOL,  { .i64 = 0    },
+      0, 1, FLAGS },
+
     { "iall",       "Set all input color properties together",
       OFFSET(user_iall),   AV_OPT_TYPE_INT, { .i64 = CS_UNSPECIFIED },
       CS_UNSPECIFIED, CS_NB - 1, FLAGS, .unit = "all" },
-- 
2.51.0.rc2.233.g662b1ed5c5-goog

Certain colors when going through the conversion can result in out of
gamut colors after the rotation. The colorspace filter allows that with
the extended range. The added clamping just keeps the colors within the
[0, 1) range rather than using that extended range. I'm not enough of a
color scientist to say which is correct, but there are certain
situations where we would prefer to keep the colors in gamut.

The example I have is:

A solid color image of 8-bit YUV: Y=157, U=164, V=98.

Specify the input as:

Input range: MPEG
In color matrix: BT470BG
In color primaries: BT470M
In color transfer characteristics: Gamma 28

Output as:
Out color range: JPEG
Out color matrix: BT.709
Out color primaries: BT.709
Out color transfer characteristics: BT.709

During the calculation you get:

Input YUV:                             y=157,      u=164,      v-98
Post-yuv2rgb BT.470BG:                 r=0.456055, g=0.684152, b=0.928606
Post-apply gamma28 linear LUT:         r=0.110979, g=0.345494, b=0.812709
Post-color rotation BT.470M to BT.709: r=-0.04161, g=0.384626, b=0.852400
Post-apply Rec.709 delinear LUT:       r=-0.16382, g=0.615932, b=0.923793
Post-rgb2yuv Rec.709 matrix:           y=120,      u=190,      v=25

Where with this change, the delinear LUT output would be clamped to 0,
so the result would be:
r=0.000000, g=0.612390, b=0.918807 and a final output of
y=129, u=185, v=46

As for the long and av_clip64, this was just because lrint returned a
long, so I left it as that and then used av_clip64 to the [0,1) range to
avoid overflow. But re-reading, it looks like av_clip_int16 would
downcast that long to int anyway so the possibility of overflow already
existed there. I've put it back to int just to match the existing
behavior. 

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

[FFmpeg-devel] Re: [PATCH] vf_colorspace: Add an option to clamp trc LUT output

[Ronald S. Bultje via ffmpeg-devel]

Hi Drew,

Thanks for the bug report!

On Mon, Aug 25, 2025 at 10:50 AM Drew Dunne via ffmpeg-devel <
ffmpeg-devel@ffmpeg.org> wrote:

> A solid color image of 8-bit YUV: Y=157, U=164, V=98.
>
> Specify the input as:
>
> Input range: MPEG
> In color matrix: BT470BG
> In color primaries: BT470M
> In color transfer characteristics: Gamma 28
>
> Output as:
> Out color range: JPEG
> Out color matrix: BT.709
> Out color primaries: BT.709
> Out color transfer characteristics: BT.709
>
> During the calculation you get:
>
> Input YUV:                             y=157,      u=164,      v-98
> Post-yuv2rgb BT.470BG:                 r=0.456055, g=0.684152, b=0.928606
> Post-apply gamma28 linear LUT:         r=0.110979, g=0.345494, b=0.812709
> Post-color rotation BT.470M to BT.709: r=-0.04161, g=0.384626, b=0.852400
> Post-apply Rec.709 delinear LUT:       r=-0.16382, g=0.615932, b=0.923793
> Post-rgb2yuv Rec.709 matrix:           y=120,      u=190,      v=25
>
> Where with this change, the delinear LUT output would be clamped to 0,
> so the result would be:
> r=0.000000, g=0.612390, b=0.918807 and a final output of
> y=129, u=185, v=46


So for those of us playing along, a repro looks like this:

$ hexdump /tmp/in.yuv
0000000 9d9d 9d9d 62a4
$ git diff
diff --git a/libavfilter/vf_colorspace.c b/libavfilter/vf_colorspace.c
index e1f4725f635..512eb620fcf 100644
--- a/libavfilter/vf_colorspace.c
+++ b/libavfilter/vf_colorspace.c
@@ -348,11 +348,20 @@ static int convert(AVFilterContext *ctx, void *data,
int job_nr, int n_jobs)
          */
         s->yuv2rgb(rgb, s->rgb_stride, in_data, td->in_linesize, w, h,
                    s->yuv2rgb_coeffs, s->yuv_offset[0]);
+        printf("post-yuv2rgb - R: %d, G: %d, B: %d\n",
+               rgb[0][0], rgb[1][0], rgb[2][0]);
         if (!s->rgb2rgb_passthrough) {
             apply_lut(rgb, s->rgb_stride, w, h, s->lin_lut);
-            if (!s->lrgb2lrgb_passthrough)
+            printf("post-linearize - R: %d, G: %d, B: %d\n",
+                   rgb[0][0], rgb[1][0], rgb[2][0]);
+            if (!s->lrgb2lrgb_passthrough) {
                 s->dsp.multiply3x3(rgb, s->rgb_stride, w, h,
s->lrgb2lrgb_coeffs);
+                printf("post-rgb2rgb - R: %d, G: %d, B: %d\n",
+                       rgb[0][0], rgb[1][0], rgb[2][0]);
+            }
             apply_lut(rgb, s->rgb_stride, w, h, s->delin_lut);
+            printf("post-delinearize - R: %d, G: %d, B: %d\n",
+                   rgb[0][0], rgb[1][0], rgb[2][0]);
         }
         if (s->dither == DITHER_FSB) {
             s->rgb2yuv_fsb(out_data, td->out_linesize, rgb, s->rgb_stride,
w, h,
$ make -j4 && ./ffmpeg -s 2x2 -pix_fmt yuv420p -f rawvideo -c:v rawvideo -i
/tmp/in.yuv -vf
colorspace=ispace=bt470bg:iprimaries=bt470m:itrc=gamma28:irange=mpeg:range=jpeg:all=bt709
-y /tmp/out.yuv
[..]
post-yuv2rgb - R: 13076, G: 19616, B: 26625
post-linearize - R: 3182, G: 9906, B: 23302
post-rgb2rgb - R: -1193, G: 11028, B: 24440
post-delinearize - R: -4697, G: 17660, B: 26487

Patch looks OK. Drew, would it be possible to submit the patch at
code.ffmpeg.org so I can merge it using our new system?

Thanks,
Ronald
_______________________________________________
ffmpeg-devel mailing list -- ffmpeg-devel@ffmpeg.org
To unsubscribe send an email to ffmpeg-devel-leave@ffmpeg.org