From 2727dbb05345204883f43f6474a5867f5a63300f Mon Sep 17 00:00:00 2001 From: Georg Martius Date: Tue, 8 Oct 2013 23:57:17 +0200 Subject: [PATCH] lavfi/vidstabdetect,vidstabtransform: update to vid.stab 0.98 In particular: * set default value for accuracy to 15 (max) * add zoomspeed for dynamic zoom (optzoom=2) * make camera path optimization algorithm configurable * update optzoom documenation in code to use new localmotions calculation function commandline arguments * add debug option in vidstabtransform Signed-off-by: Georg Martius Signed-off-by: Stefano Sabatini --- doc/filters.texi | 32 ++++++++++++++++++++++++----- libavfilter/vf_vidstabdetect.c | 3 ++- libavfilter/vf_vidstabtransform.c | 34 +++++++++++++++++++++++++------ 3 files changed, 57 insertions(+), 12 deletions(-) diff --git a/doc/filters.texi b/doc/filters.texi index 0f9036535d..64dd084a56 100644 --- a/doc/filters.texi +++ b/doc/filters.texi @@ -8064,7 +8064,7 @@ value of 10 means strong shakiness. Default value is 5. @item accuracy Set the accuracy of the detection process. It must be a value in the range 1-15. A value of 1 means low accuracy, a value of 15 means high -accuracy. Default value is 9. +accuracy. Default value is 15. @item stepsize Set stepsize of the search process. The region around minimum is @@ -8144,11 +8144,21 @@ This filter accepts the following options: path to the file used to read the transforms (default: @file{transforms.trf}) @item smoothing -number of frames (value*2 + 1) used for lowpass filtering the camera movements +Set the number of frames (value*2 + 1) used for lowpass filtering the camera movements (default: 10). For example a number of 10 means that 21 frames are used (10 in the past and 10 in the future) to smoothen the motion in the video. A larger values leads to a smoother video, but limits the acceleration of the camera (pan/tilt movements). +0 is a special case where a static camera is simulated. + +@item optalgo +Set the camera path optimization algorithm: +@table @samp +@item gauss +gaussian kernel low-pass filter on camera motion (default) +@item avg +averaging on transformations +@end table @item maxshift maximal number of pixels to translate frames (default: -1 no limit) @@ -8186,7 +8196,7 @@ relative to previous frame (default) @end table @item zoom -percentage to zoom (default: 0) +Set percentage to zoom (default: 0) @table @samp @item >0 zoom in @@ -8195,18 +8205,21 @@ zoom out @end table @item optzoom -set optimal zooming to avoid borders +Set optimal zooming to avoid borders @table @samp @item 0 disabled @item 1 optimal static zoom value is determined (only very strong movements will lead to visible borders) (default) @item 2 -optimal adaptive zoom value is determined (no borders will be visible) +optimal adaptive zoom value is determined (no borders will be visible), see @option{zoomspeed} @end table Note that the value given at zoom is added to the one calculated here. +@item zoomspeed +Set percent to zoom maximally each frame (for @option{optzoom=2}). Range is from 0 to 5, default value is 0.2 + @item interpol type of interpolation @@ -8233,6 +8246,15 @@ off (default) virtual tripod mode: equivalent to @code{relative=0:smoothing=0} @end table +@item debug +Increase log verbosity of set to 1. Also the detected global motions are written to the temporary file @file{global_motions.trf}. +@table @samp +@item 0 +disabled (default) +@item 1 +enabled +@end table + @end table @subsection Examples diff --git a/libavfilter/vf_vidstabdetect.c b/libavfilter/vf_vidstabdetect.c index b2977ff730..008c993368 100644 --- a/libavfilter/vf_vidstabdetect.c +++ b/libavfilter/vf_vidstabdetect.c @@ -49,7 +49,7 @@ static const AVOption vidstabdetect_options[] = { {"result", "path to the file used to write the transforms", OFFSET(result), AV_OPT_TYPE_STRING, {.str = DEFAULT_RESULT_NAME}, .flags = FLAGS}, {"shakiness", "how shaky is the video and how quick is the camera?" " 1: little (fast) 10: very strong/quick (slow)", OFFSETC(shakiness), AV_OPT_TYPE_INT, {.i64 = 5}, 1, 10, FLAGS}, - {"accuracy", "(>=shakiness) 1: low 15: high (slow)", OFFSETC(accuracy), AV_OPT_TYPE_INT, {.i64 = 9}, 1, 15, FLAGS}, + {"accuracy", "(>=shakiness) 1: low 15: high (slow)", OFFSETC(accuracy), AV_OPT_TYPE_INT, {.i64 = 15}, 1, 15, FLAGS}, {"stepsize", "region around minimum is scanned with 1 pixel resolution", OFFSETC(stepSize), AV_OPT_TYPE_INT, {.i64 = 6}, 1, 32, FLAGS}, {"mincontrast", "below this contrast a field is discarded (0-1)", OFFSETC(contrastThreshold), AV_OPT_TYPE_DOUBLE, {.dbl = 0.25}, 0.0, 1.0, FLAGS}, {"show", "0: draw nothing; 1,2: show fields and transforms", OFFSETC(show), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 2, FLAGS}, @@ -135,6 +135,7 @@ static int config_input(AVFilterLink *inlink) av_log(ctx, AV_LOG_INFO, " accuracy = %d\n", sd->conf.accuracy); av_log(ctx, AV_LOG_INFO, " stepsize = %d\n", sd->conf.stepSize); av_log(ctx, AV_LOG_INFO, " mincontrast = %f\n", sd->conf.contrastThreshold); + av_log(ctx, AV_LOG_INFO, " tripod = %d\n", sd->conf.virtualTripod); av_log(ctx, AV_LOG_INFO, " show = %d\n", sd->conf.show); av_log(ctx, AV_LOG_INFO, " result = %s\n", sd->result); diff --git a/libavfilter/vf_vidstabtransform.c b/libavfilter/vf_vidstabtransform.c index 3c0a5ee40d..b4ef1c225f 100644 --- a/libavfilter/vf_vidstabtransform.c +++ b/libavfilter/vf_vidstabtransform.c @@ -39,6 +39,7 @@ typedef struct { VSTransformations trans; // transformations char *input; // name of transform file int tripod; + int debug; } TransformContext; #define OFFSET(x) offsetof(TransformContext, x) @@ -49,7 +50,15 @@ static const AVOption vidstabtransform_options[] = { {"input", "path to the file storing the transforms", OFFSET(input), AV_OPT_TYPE_STRING, {.str = DEFAULT_INPUT_NAME}, .flags = FLAGS }, {"smoothing", "number of frames*2 + 1 used for lowpass filtering", OFFSETC(smoothing), - AV_OPT_TYPE_INT, {.i64 = 10}, 1, 1000, FLAGS}, + AV_OPT_TYPE_INT, {.i64 = 15}, 0, 1000, FLAGS}, + {"optalgo", "camera path optimization algo", OFFSETC(camPathAlgo), + AV_OPT_TYPE_INT, {.i64 = VSOptimalL1}, VSOptimalL1, VSAvg, FLAGS, "optalgo"}, + { "opt", "global optimization", 0, // from version 1.0 on + AV_OPT_TYPE_CONST, {.i64 = VSOptimalL1 }, 0, 0, FLAGS, "optalgo"}, + { "gauss", "gaussian kernel", 0, + AV_OPT_TYPE_CONST, {.i64 = VSGaussian }, 0, 0, FLAGS, "optalgo"}, + { "avg", "simple averaging on motion", 0, + AV_OPT_TYPE_CONST, {.i64 = VSAvg }, 0, 0, FLAGS, "optalgo"}, {"maxshift", "maximal number of pixels to translate image", OFFSETC(maxShift), AV_OPT_TYPE_INT, {.i64 = -1}, -1, 500, FLAGS}, {"maxangle", "maximal angle in rad to rotate image", OFFSETC(maxAngle), @@ -66,8 +75,10 @@ static const AVOption vidstabtransform_options[] = { AV_OPT_TYPE_INT, {.i64 = 1}, 0, 1, FLAGS}, {"zoom", "percentage to zoom >0: zoom in, <0 zoom out", OFFSETC(zoom), AV_OPT_TYPE_DOUBLE, {.dbl = 0}, -100, 100, FLAGS}, - {"optzoom", "0: nothing, 1: determine optimal zoom (added to 'zoom')", OFFSETC(optZoom), + {"optzoom", "0: nothing, 1: optimal static zoom, 2: optimal dynamic zoom", OFFSETC(optZoom), AV_OPT_TYPE_INT, {.i64 = 1}, 0, 2, FLAGS}, + {"zoomspeed", "for adative zoom: percent to zoom maximally each frame", OFFSETC(zoomSpeed), + AV_OPT_TYPE_DOUBLE, {.dbl = 0.25}, 0, 5, FLAGS}, {"interpol", "type of interpolation", OFFSETC(interpolType), AV_OPT_TYPE_INT, {.i64 = 2}, 0, 3, FLAGS, "interpol"}, { "no", "no interpolation", 0, @@ -80,6 +91,8 @@ static const AVOption vidstabtransform_options[] = { AV_OPT_TYPE_CONST, {.i64 = VS_BiCubic },0, 0, FLAGS, "interpol"}, {"tripod", "if 1: virtual tripod mode (equiv. to relative=0:smoothing=0)", OFFSET(tripod), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, FLAGS}, + {"debug", "if 1: more output printed and global motions are stored to file",OFFSET(debug), + AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, FLAGS}, {NULL} }; @@ -153,12 +166,15 @@ static int config_input(AVFilterLink *inlink) // set values that are not initializes by the options tc->conf.modName = "vidstabtransform"; - tc->conf.verbose =1; + tc->conf.verbose = 1 + tc->debug; if (tc->tripod) { - av_log(ctx, AV_LOG_INFO, "Virtual tripod mode: relative=0, smoothing=0"); + av_log(ctx, AV_LOG_INFO, "Virtual tripod mode: relative=0, smoothing=0\n"); tc->conf.relative = 0; tc->conf.smoothing = 0; } + tc->conf.simpleMotionCalculation = 0; + tc->conf.storeTransforms = tc->debug; + tc->conf.smoothZoom = 0; if (vsTransformDataInit(td, &tc->conf, &fi_src, &fi_dest) != VS_OK) { av_log(ctx, AV_LOG_ERROR, "initialization of vid.stab transform failed, please report a BUG\n"); @@ -169,13 +185,19 @@ static int config_input(AVFilterLink *inlink) av_log(ctx, AV_LOG_INFO, "Video transformation/stabilization settings (pass 2/2):\n"); av_log(ctx, AV_LOG_INFO, " input = %s\n", tc->input); av_log(ctx, AV_LOG_INFO, " smoothing = %d\n", tc->conf.smoothing); + av_log(ctx, AV_LOG_INFO, " optalgo = %s\n", + tc->conf.camPathAlgo == VSOptimalL1 ? "opt" : + (tc->conf.camPathAlgo == VSGaussian ? "gauss" : "avg" )); av_log(ctx, AV_LOG_INFO, " maxshift = %d\n", tc->conf.maxShift); av_log(ctx, AV_LOG_INFO, " maxangle = %f\n", tc->conf.maxAngle); av_log(ctx, AV_LOG_INFO, " crop = %s\n", tc->conf.crop ? "Black" : "Keep"); av_log(ctx, AV_LOG_INFO, " relative = %s\n", tc->conf.relative ? "True": "False"); av_log(ctx, AV_LOG_INFO, " invert = %s\n", tc->conf.invert ? "True" : "False"); av_log(ctx, AV_LOG_INFO, " zoom = %f\n", tc->conf.zoom); - av_log(ctx, AV_LOG_INFO, " optzoom = %s\n", tc->conf.optZoom ? "On" : "Off"); + av_log(ctx, AV_LOG_INFO, " optzoom = %s\n", + tc->conf.optZoom == 1 ? "Static (1)" : (tc->conf.optZoom == 2 ? "Dynamic (2)" : "Off (0)" )); + if (tc->conf.optZoom == 2) + av_log(ctx, AV_LOG_INFO, " zoomspeed = %g\n", tc->conf.zoomSpeed ); av_log(ctx, AV_LOG_INFO, " interpol = %s\n", getInterpolationTypeName(tc->conf.interpolType)); f = fopen(tc->input, "r"); @@ -186,7 +208,7 @@ static int config_input(AVFilterLink *inlink) VSManyLocalMotions mlms; if (vsReadLocalMotionsFile(f, &mlms) == VS_OK) { // calculate the actual transforms from the local motions - if (vsLocalmotions2TransformsSimple(td, &mlms, &tc->trans) != VS_OK) { + if (vsLocalmotions2Transforms(td, &mlms, &tc->trans) != VS_OK) { av_log(ctx, AV_LOG_ERROR, "calculating transformations failed\n"); return AVERROR(EINVAL); }