[FFmpeg-devel] rectification filter
Daniel Oberhoff
danieloberhoff at gmail.com
Sun Aug 3 18:43:18 CEST 2014
Am 03.08.2014 um 18:38 schrieb Paul B Mahol <onemda at gmail.com>:
> On Sun, Aug 3, 2014 at 5:39 PM, Daniel Oberhoff <danieloberhoff at gmail.com>
> wrote:
>
>>
>> Am 03.08.2014 um 03:15 schrieb Daniel Oberhoff <
>> danieloberhoff at googlemail.com>:
>>
>>> Am 03.08.2014 um 00:12 schrieb Clément Bœsch <u at pkh.me>:
>>>
>>>> On Fri, Aug 01, 2014 at 12:56:42PM +0200, Daniel Oberhoff wrote:
>>>>>
>>>>> Am 01.08.2014 um 12:22 schrieb Clément Bœsch <u at pkh.me>:
>>>>>
>>>>>> On Fri, Aug 01, 2014 at 12:13:22PM +0200, Daniel Oberhoff wrote:
>>>>>>>
>>>>>>> Am 29.07.2014 um 09:54 schrieb Carl Eugen Hoyos <cehoyos at ag.or.at>:
>>>>>>>
>>>>>>>> Daniel Oberhoff <danieloberhoff <at> gmail.com> writes:
>>>>>>>>
>>>>>>>>> OBJS-$(CONFIG_ZOOMPAN_FILTER) += vf_zoompan.o
>>>>>>>>> +OBJS-$(CONFIG_RECTIFICATION_FILTER) += vf_rectification.o
>>>>>>>>
>>>>>>>>> REGISTER_FILTER(ZOOMPAN, zoompan, vf);
>>>>>>>>> + REGISTER_FILTER(RECTIFICATION, rectification, vf);
>>>>>>>>
>>>>>>>> Keep the alphabetic ordering please.
>>>>>>>>
>>>>>>>
>>>>>>> Ok
>>>>>>>
>>>>>>>>> + AV_PIX_FMT_YUV410P,
>>>>>>>>> + AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUVJ444P,
>>>>>>>>> + AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUVJ420P,
>>>>>>>>> + AV_PIX_FMT_YUVA444P, AV_PIX_FMT_YUVA420P,
>>>>>>>>> + AV_PIX_FMT_NONE
>>>>>>>>
>>>>>>>> I have no idea what this filter does and since the
>>>>>>>> most important format (yuv420p) is supported, this
>>>>>>>> doesn't really matter but yuv410p and yuva444p are
>>>>>>>> quite exotic formats, I would at least have
>>>>>>>> expected yuv422p, yuva420p and yuv444p in the list.
>>>>>>>>
>>>>>>>> Did you test all above formats (at least the left row)?
>>>>>>>
>>>>>>> Added the one from your list that was missing. Can you suggest an
>> easy way of testing all these formats?
>>>>>>>
>>>>>>
>>>>>>
>>>>>> add a format filter before yours in the filtergraph (-vf
>>>>>> format=yuv410p,rectification …)
>>>>>
>>>>> Ok, borders can be a bitt different, but otherwise all work now,
>> except vuy444p, which ffmpeg reports as invalid. btw, how can I get the
>> components of a given color in the colorspace?
>>>>>
>>>>>> If you want to add a FATE test (which would be really awesome), you
>> can
>>>>>> grep for "video_filter" in tests/fate/filter-video.mak (to test all
>> pixel
>>>>>> formats the filter supports), or just make a standard test with
>> framecrc.
>>>>>>
>>>>>> [...]
>>>>>
>>>>> I did that by adding this:
>>>>>
>>>>> FATE_FILTER_PIXFMTS-$(CONFIG_RECTIFICATION_FILTER) +=
>> fate-filter-pixfmts-rectification
>>>>> fate-filter-pixfmts-rectification: CMD = pixfmts "0.6:0.4:0.65:0.4"
>>>>>
>>>>> then I ran
>>>>>
>>>>> make fate-filter-pixfmts-rectification GEN=1
>>>>>
>>>>> which made a file
>>>>>
>>>>> test/data/fate/filter-pixfmts-rectification.rep
>>>>>
>>>>
>>>> This is the result of a normal run to be compared by with the reference
>>>> which should have also been generated with the help of GEN=1 for your
>>>> first run. The file(s) generated by GEN=1 need to be tracked.
>>>>
>>>>> do I check this in or not?
>>>>>
>>>>
>>>> Of course you are supposed to check if the results make sense.
>>>>
>>>> Use make fate-filter-pixfmts-rectifications V=1 to see what's going on.
>>>>
>>>>> running
>>>>>
>>>>> make fate-filter-pixfmts-rectification GEN=1
>>>>>
>>>>> runs fine…
>>>>>
>>>>
>>>> Of course, it doesn't do the comparison, it generates the reference.
>>>>
>>>>> Best
>>>>>
>>>>> Daniel
>>>>
>>>> --
>>>> Clément B.
>>>> _______________________________________________
>>>> ffmpeg-devel mailing list
>>>> ffmpeg-devel at ffmpeg.org
>>>> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>>>
>>> Alright, will get that cleaned up. Since I seem to have all consent I
>> need to also change the license I will see that I get a new patch ready
>> soon.
>>>
>>> Best and thanks for feedback!
>>>
>>> Daniel
>>
>> Hello all,
>>
>> I updated the patch trying to incorporate all review feedback. I also got
>> consent from the original author to put this filter under LGPL, and thus
>> have it compiled in by default. I also conversed with Cyrille from Krita
>> and he doesn’t see any more copyright issues, as at that point it boils
>> down to the use of a well known algorithm, in a straight-forward
>> implementation.
>>
>> Compared to the previous patch I renamed the filter to lenscorrection,
>> since that is the name in the frei0r suite, and it will make using this
>> instead of the frei0r one as easy as stating
>>
>> lenscorrection=...
>>
>> instead of
>>
>> frei0r=lenscorrection:...
>>
>> Last but not least I adapted the format list, inspected results on all
>> formats, and supplied a fate test. Release notes and docs where also
>> adapted. Please check if this is ok to push now.
>>
>> From dc552ae06a41725988250896327af2cceee1b812 Mon Sep 17 00:00:00 2001
>> From: Daniel Oberhoff <daniel at danieloberhoff.de>
>> Date: Mon, 28 Jul 2014 23:58:12 +0200
>> Subject: [PATCH] ported lenscorrection filter from frei0r
>>
>> ---
>> Changelog | 2 +-
>> doc/filters.texi | 36 +++++
>> libavfilter/Makefile | 1 +
>> libavfilter/allfilters.c | 1 +
>> libavfilter/version.h | 4 +-
>> libavfilter/vf_lenscorrection.c | 208
>> +++++++++++++++++++++++++++
>> tests/fate/filter-video.mak | 3 +
>> tests/ref/fate/filter-pixfmts-lenscorrection | 8 ++
>> 8 files changed, 260 insertions(+), 3 deletions(-)
>> create mode 100644 libavfilter/vf_lenscorrection.c
>> create mode 100644 tests/ref/fate/filter-pixfmts-lenscorrection
>>
>>
>>
>> diff --git a/libavfilter/Makefile b/libavfilter/Makefile
>> index 0f54381..f74defa 100644
>> --- a/libavfilter/Makefile
>> +++ b/libavfilter/Makefile
>> @@ -162,6 +162,7 @@ OBJS-$(CONFIG_PIXDESCTEST_FILTER) +=
>> vf_pixdesctest.o
>> OBJS-$(CONFIG_PP_FILTER) += vf_pp.o
>> OBJS-$(CONFIG_PSNR_FILTER) += vf_psnr.o dualinput.o
>> framesync.o
>> OBJS-$(CONFIG_PULLUP_FILTER) += vf_pullup.o
>> +OBJS-$(CONFIG_LENSCORRECTION_FILTER) += vf_lenscorrection.o
>>
>
> not in alphabetical order
>
> E, {.dbl=0.5}, 0, 1, .flags=FLAGS },
>> + { "k1", "set quadratic distortion factor",
>> offsetof(LenscorrectionCtx, k1), AV_OPT_TYPE_DOUBLE, {.dbl=0.5}, 0, 1,
>> .flags=FLAGS },
>> + { "k2", "set double quadratic distortion factor",
>> offsetof(LenscorrectionCtx, k2), AV_OPT_TYPE_DOUBLE, {.dbl=0.5}, 0, 1,
>> .flags=FLAGS },
>> + { NULL }
>> +};
>> +
>> +AVFILTER_DEFINE_CLASS(Lenscorrection);
>> +
>> +static av_cold int init(AVFilterContext *ctx)
>> +{
>> + return 0;
>> +}
>>
>
> please remove this one, there is no point in having code that does nothing.
>
>> +
>> +static av_cold void uninit(AVFilterContext *ctx)
>> +{
>> +}
>> +
>>
>
> ditto
>
Done thanks!
>
>
>> +typedef struct ThreadData {
>> + AVFrame *in, *out;
>> + float w, h;
>> + int plane;
>> + float xcenter, ycenter;
>> + float k1, k2;
>> +} ThreadData;
>> +
>> +static int filter_slice(AVFilterContext *ctx, void *arg, int job, int
>> nb_jobs)
>> +{
>> + ThreadData *td = (ThreadData*)arg;
>> + AVFrame *in = td->in;
>> + AVFrame *out = td->out;
>> +
>> + const float w = td->w, h = td->h;
>> + const float xcenter = td->xcenter;
>> + const float ycenter = td->ycenter;
>> + const float r2inv = 4.0 / (w * w + h * h);
>> + const float k1 = td->k1 - 0.5;
>> + const float k2 = td->k2 - 0.5;
>> + const int start = (h * job ) / nb_jobs;
>> + const int end = (h * (job+1)) / nb_jobs;
>> + const int plane = td->plane;
>> + const int inlinesize = in->linesize[plane];
>> + const int outlinesize = out->linesize[plane];
>> + const uint8_t *indata = in->data[plane];
>> + uint8_t *outrow = out->data[plane] + start * outlinesize;
>> + int i;
>> + for (i = start; i < end; i++, outrow += outlinesize) {
>> + const float off_y = i - ycenter;
>> + const float off_y2 = off_y * off_y;
>> + uint8_t *out = outrow;
>> + int j;
>> + for (j = 0; j < w; j++) {
>> + const float off_x = j - xcenter;
>> + const float r2 = (off_x * off_x + off_y2) * r2inv;
>> + const float radius_mult = 1.0f + r2 * k1 + r2 * r2 * k2;
>> + const int x = xcenter + radius_mult * off_x + 0.5f;
>> + const int y = ycenter + radius_mult * off_y + 0.5f;
>> + const char isvalid = x > 0 && x < w - 1 && y > 0 && y < h - 1;
>> + *out++ = isvalid ? indata[y * inlinesize + x] : 0;
>> + }
>> + }
>> + return 0;
>> +}
>> +
>> +static int query_formats(AVFilterContext *ctx)
>> +{
>> + static enum PixelFormat pix_fmts[] = {
>> + AV_PIX_FMT_YUV410P,
>> + AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUVJ444P,
>> + AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUVJ420P,
>> + AV_PIX_FMT_YUVA444P, AV_PIX_FMT_YUVA420P,
>> + AV_PIX_FMT_YUV422P,
>> + AV_PIX_FMT_NONE
>> + };
>> +
>> + ff_set_common_formats(ctx, ff_make_format_list(pix_fmts));
>> + return 0;
>> +}
>> +
>> +static int config_props(AVFilterLink *outlink)
>> +{
>> + AVFilterContext* ctx = outlink->src;
>> + LenscorrectionCtx* rect = ctx->priv;
>> + AVFilterLink *inlink = ctx->inputs[0];
>> + const AVPixFmtDescriptor *pixdesc =
>> av_pix_fmt_desc_get(inlink->format);
>> + rect->hsub = pixdesc->log2_chroma_w;
>> + rect->vsub = pixdesc->log2_chroma_h;
>> + outlink->w = rect->width = inlink->w;
>> + outlink->h = rect->height = inlink->h;
>> + rect->nb_planes = av_pix_fmt_count_planes(inlink->format);
>> + return 0;
>> +}
>> +
>> +static int filter_frame(AVFilterLink *inlink, AVFrame *in)
>> +{
>> + AVFilterContext *ctx = inlink->dst;
>> + AVFilterLink *outlink = ctx->outputs[0];
>> + LenscorrectionCtx *rect = (LenscorrectionCtx*)ctx->priv;
>> + AVFrame* out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
>> + int plane;
>> +
>> + if (!out) {
>> + av_frame_free(&in);
>> + return AVERROR(ENOMEM);
>> + }
>> +
>> + av_frame_copy_props(out, in);
>> +
>> + for (plane = 0; plane < rect->nb_planes; ++plane) {
>> + int hsub = plane == 1 || plane == 2 ? rect->hsub : 0;
>> + int vsub = plane == 1 || plane == 2 ? rect->vsub : 0;
>> + float hdiv = 1 << hsub;
>> + float vdiv = 1 << vsub;
>> + float w = rect->width / hdiv;
>> + float h = rect->height / vdiv;
>> + ThreadData td = { .in = in, .out = out,
>> + .w = w,
>> + .h = h,
>> + .xcenter = rect->cx * w,
>> + .ycenter = rect->cy * h,
>> + .k1 = rect->k1,
>> + .k2 = rect->k2,
>> + .plane = plane};
>> + ctx->internal->execute(ctx, filter_slice, &td, NULL, FFMIN(h,
>> ctx->graph->nb_threads));
>> + }
>> +
>> + av_frame_free(&in);
>> + return ff_filter_frame(outlink, out);
>> +}
>> +
>> +static const AVFilterPad Lenscorrection_inputs[] = {
>> + {
>> + .name = "default",
>> + .type = AVMEDIA_TYPE_VIDEO,
>> + .filter_frame = filter_frame,
>> + },
>> + { NULL }
>> +};
>> +
>> +static const AVFilterPad Lenscorrection_outputs[] = {
>> + {
>> + .name = "default",
>> + .type = AVMEDIA_TYPE_VIDEO,
>> + .config_props = config_props,
>> + },
>> + { NULL }
>> +};
>> +
>> +AVFilter ff_vf_lenscorrection = {
>> + .name = "lenscorrection",
>> + .description = NULL_IF_CONFIG_SMALL("Rectify the image by
>> correcting for lens distortion."),
>> + .priv_size = sizeof(LenscorrectionCtx),
>> + .init = init,
>> + .uninit = uninit,
>>
>
> Please remove those 2 lines.
>
>
>> + .query_formats = query_formats,
>> + .inputs = Lenscorrection_inputs,
>> + .outputs = Lenscorrection_outputs,
>> + .priv_class = &Lenscorrection_class,
>> + .flags = AVFILTER_FLAG_SLICE_THREADS,
>> +};
>> +
>> \ No newline at end of file
>> diff --git a/tests/fate/filter-video.mak b/tests/fate/filter-video.mak
>> index d87129b..cc5a71b 100644
>> --- a/tests/fate/filter-video.mak
>> +++ b/tests/fate/filter-video.mak
>> @@ -331,6 +331,9 @@ fate-filter-pixfmts-il: CMD = pixfmts
>> "luma_mode=d:chroma_mode=d:alpha_mode=d
>> FATE_FILTER_PIXFMTS-$(CONFIG_KERNDEINT_FILTER) +=
>> fate-filter-pixfmts-kerndeint
>> fate-filter-pixfmts-kerndeint: CMD = pixfmts ""
>> "tinterlace=interleave_top,"
>>
>> +FATE_FILTER_PIXFMTS-$(CONFIG_LENSCORRECTION_FILTER) +=
>> fate-filter-pixfmts-lenscorrection
>> +fate-filter-pixfmts-lenscorrection: CMD = pixfmts "0.6:0.4:0.65:0.4"
>> +
>> FATE_FILTER_PIXFMTS-$(CONFIG_LUT_FILTER) += fate-filter-pixfmts-lut
>> fate-filter-pixfmts-lut: CMD = pixfmts
>> "c0=2*val:c1=2*val:c2=val/2:c3=negval+40"
>>
>> diff --git a/tests/ref/fate/filter-pixfmts-lenscorrection
>> b/tests/ref/fate/filter-pixfmts-lenscorrection
>> new file mode 100644
>> index 0000000..18af6fb
>> --- /dev/null
>> +++ b/tests/ref/fate/filter-pixfmts-lenscorrection
>> @@ -0,0 +1,8 @@
>> +yuv410p e7d59dbdb1afab7e2a8f770d563e28c4
>> +yuv420p bc45b4762d5271410ff825317c85af64
>> +yuv422p 5cce0c299322634d65e6b32c976e2c12
>> +yuv444p 03de9a93ab3045a523b234ea93f21c91
>> +yuva420p d1fa6735c4e7fbbf3a501cec1f0b4ac1
>> +yuva444p 713ddf5861d3df11c70a242a13c5e92e
>> +yuvj420p 1d5cccaf4ef568ae9fa36f9a28e71c34
>> +yuvj444p aef1db29848e3b1dcaf4309255c38cbd
>> --
>> 1.8.2
>>
>> _______________________________________________
>> ffmpeg-devel mailing list
>> ffmpeg-devel at ffmpeg.org
>> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>>
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
More information about the ffmpeg-devel
mailing list