diff mbox

[FFmpeg-devel] avfilter: add ADM filter

Message ID 1501239185-4593-1-git-send-email-ashk43712@gmail.com
State Superseded
Headers show

Commit Message

Ashish Singh July 28, 2017, 10:53 a.m. UTC
From: Ashish Singh <ashk43712@gmail.com>

Signed-off-by: Ashish Singh <ashk43712@gmail.com>
---
 Changelog                |   1 +
 doc/filters.texi         |  19 ++
 libavfilter/Makefile     |   1 +
 libavfilter/adm.h        |  88 +++++
 libavfilter/allfilters.c |   1 +
 libavfilter/vf_adm.c     | 819 +++++++++++++++++++++++++++++++++++++++++++++++
 6 files changed, 929 insertions(+)
 create mode 100644 libavfilter/adm.h
 create mode 100644 libavfilter/vf_adm.c
diff mbox

Patch

diff --git a/Changelog b/Changelog
index 187ae79..5a9b211 100644
--- a/Changelog
+++ b/Changelog
@@ -29,6 +29,7 @@  version <next>:
 - limiter video filter
 - libvmaf video filter
 - Dolby E decoder and SMPTE 337M demuxer
+- adm video filter
 
 version 3.3:
 - CrystalHD decoder moved to new decode API
diff --git a/doc/filters.texi b/doc/filters.texi
index 2324b96..86cf865 100644
--- a/doc/filters.texi
+++ b/doc/filters.texi
@@ -4604,6 +4604,25 @@  build.
 
 Below is a description of the currently available video filters.
 
+@section adm
+
+Obtain the average ADM/DLM (Detail Loss Metric) between two input videos.
+
+This filter takes two input videos.
+
+Both input videos must have the same resolution and pixel format for
+this filter to work correctly. Also it assumes that both inputs
+have the same number of frames, which are compared one by one.
+
+The obtained average ADM is printed through the logging system.
+
+In the below example the input file @file{main.mpg} being processed is compared
+with the reference file @file{ref.mpg}.
+
+@example
+ffmpeg -i main.mpg -i ref.mpg -lavfi adm -f null -
+@end example
+
 @section alphaextract
 
 Extract the alpha component from the input as a grayscale video. This
diff --git a/libavfilter/Makefile b/libavfilter/Makefile
index ee16361..bca3eff 100644
--- a/libavfilter/Makefile
+++ b/libavfilter/Makefile
@@ -126,6 +126,7 @@  OBJS-$(CONFIG_SINE_FILTER)                   += asrc_sine.o
 OBJS-$(CONFIG_ANULLSINK_FILTER)              += asink_anullsink.o
 
 # video filters
+OBJS-$(CONFIG_ADM_FILTER)                    += vf_adm.o dualinput.o framesync.o
 OBJS-$(CONFIG_ALPHAEXTRACT_FILTER)           += vf_extractplanes.o
 OBJS-$(CONFIG_ALPHAMERGE_FILTER)             += vf_alphamerge.o
 OBJS-$(CONFIG_ASS_FILTER)                    += vf_subtitles.o
diff --git a/libavfilter/adm.h b/libavfilter/adm.h
new file mode 100644
index 0000000..8b2db8b
--- /dev/null
+++ b/libavfilter/adm.h
@@ -0,0 +1,88 @@ 
+/*
+ * Copyright (c) 2017 Ronald S. Bultje <rsbultje@gmail.com>
+ * Copyright (c) 2017 Ashish Pratap Singh <ashk43712@gmail.com>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#ifndef AVFILTER_ADM_H
+#define AVFILTER_ADM_H
+
+#define VIEW_DIST 3.0f
+#define REF_DISPLAY_HEIGHT 1080
+/** Percentage of frame to discard on all 4 sides */
+#define ADM_BORDER_FACTOR (0.1)
+
+typedef struct adm_dwt_band_t {
+    float *band_a; /** Low-pass V + low-pass H. */
+    float *band_v; /** Low-pass V + high-pass H. */
+    float *band_h; /** High-pass V + low-pass H. */
+    float *band_d; /** High-pass V + high-pass H. */
+} adm_dwt_band_t;
+
+/**
+ * The following dwt visibility threshold parameters are taken from
+ * "Visibility of Wavelet Quantization Noise"
+ * by A. B. Watson, G. Y. Yang, J. A. Solomon and J. Villasenor
+ * IEEE Trans. on Image Processing, Vol. 6, No 8, Aug. 1997
+ * Page 1170, formula (7) and corresponding Table IV
+ * Table IV has 2 entries for Cb and Cr thresholds
+ * Chose those corresponding to subject "sfl" since they are lower
+ * These thresholds were obtained and modeled for the 7-9 biorthogonal wavelet basis
+ */
+struct dwt_model_params {
+    float a;
+    float k;
+    float f0;
+    float g[4];
+};
+
+static const float dwt2_db2_coeffs_lo[4] = {  0.482962913144690,  0.836516303737469, 0.224143868041857, -0.129409522550921 };
+static const float dwt2_db2_coeffs_hi[4] = { -0.129409522550921, -0.224143868041857, 0.836516303737469, -0.482962913144690 };
+
+/** 0 -> Y, 1 -> Cb, 2 -> Cr */
+static const struct dwt_model_params dwt_7_9_YCbCr_threshold[3] = {
+    { .a = 0.495, .k = 0.466, .f0 = 0.401, .g = { 1.501, 1.0, 0.534, 1.0} },
+    { .a = 1.633, .k = 0.353, .f0 = 0.209, .g = { 1.520, 1.0, 0.502, 1.0} },
+    { .a = 0.944, .k = 0.521, .f0 = 0.404, .g = { 1.868, 1.0, 0.516, 1.0} }
+};
+
+/**
+ * The following dwt basis function amplitudes, A(lambda,theta), are taken from
+ * "Visibility of Wavelet Quantization Noise"
+ * by A. B. Watson, G. Y. Yang, J. A. Solomon and J. Villasenor
+ * IEEE Trans. on Image Processing, Vol. 6, No 8, Aug. 1997
+ * Page 1172, Table V
+ * The table has been transposed, i.e. it can be used directly to obtain A[lambda][theta]
+ * These amplitudes were calculated for the 7-9 biorthogonal wavelet basis
+ */
+static const float dwt_7_9_basis_function_amplitudes[6][4] = {
+    { 0.62171,  0.67234,  0.72709,  0.67234  },
+    { 0.34537,  0.41317,  0.49428,  0.41317  },
+    { 0.18004,  0.22727,  0.28688,  0.22727  },
+    { 0.091401, 0.11792,  0.15214,  0.11792  },
+    { 0.045943, 0.059758, 0.077727, 0.059758 },
+    { 0.023013, 0.030018, 0.039156, 0.030018 }
+};
+
+/** function to compute adm score */
+int compute_adm2(const float *ref, const float *main, int w, int h,
+                 int ref_stride, int main_stride, double *score,
+                 double *score_num, double *score_den, double *scores,
+                 float *data_buf, float *temp_lo, float* temp_hi);
+
+#endif /* AVFILTER_ADM_H */
diff --git a/libavfilter/allfilters.c b/libavfilter/allfilters.c
index b1c2d11..bbd7db8 100644
--- a/libavfilter/allfilters.c
+++ b/libavfilter/allfilters.c
@@ -138,6 +138,7 @@  static void register_all(void)
 
     REGISTER_FILTER(ANULLSINK,      anullsink,      asink);
 
+    REGISTER_FILTER(ADM,            adm,            vf);
     REGISTER_FILTER(ALPHAEXTRACT,   alphaextract,   vf);
     REGISTER_FILTER(ALPHAMERGE,     alphamerge,     vf);
     REGISTER_FILTER(ASS,            ass,            vf);
diff --git a/libavfilter/vf_adm.c b/libavfilter/vf_adm.c
new file mode 100644
index 0000000..0c3effc
--- /dev/null
+++ b/libavfilter/vf_adm.c
@@ -0,0 +1,819 @@ 
+/*
+ * Copyright (c) 2017 Ronald S. Bultje <rsbultje@gmail.com>
+ * Copyright (c) 2017 Ashish Pratap Singh <ashk43712@gmail.com>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+/**
+ * @file
+ * Calculate the ADM between two input videos.
+ */
+
+#include "libavutil/avstring.h"
+#include "libavutil/opt.h"
+#include "libavutil/pixdesc.h"
+#include "avfilter.h"
+#include "dualinput.h"
+#include "drawutils.h"
+#include "formats.h"
+#include "internal.h"
+#include "adm.h"
+#include "video.h"
+#include <emmintrin.h>
+
+typedef struct ADMContext {
+    const AVClass *class;
+    FFDualInputContext dinput;
+    const AVPixFmtDescriptor *desc;
+    int width;
+    int height;
+    float *ref_data;
+    float *main_data;
+    float *data_buf;
+    float *temp_lo;
+    float *temp_hi;
+    double adm_sum;
+    uint64_t nb_frames;
+} ADMContext;
+
+static const AVOption adm_options[] = {
+    { NULL }
+};
+
+AVFILTER_DEFINE_CLASS(adm);
+
+#define MAX_ALIGN 32
+#define ALIGN_CEIL(x) ((x) + ((x) % MAX_ALIGN ? MAX_ALIGN - (x) % MAX_ALIGN : 0))
+#define OPT_RANGE_PIXEL_OFFSET (-128)
+#define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
+
+static float rcp(float x)
+{
+    float xi = _mm_cvtss_f32(_mm_rcp_ss(_mm_load_ss(&x)));
+    return xi + xi * (1.0f - x * xi);
+}
+
+#define DIVS(n, d) ((n) * rcp(d))
+
+/**
+ * lambda = 0 (finest scale), 1, 2, 3 (coarsest scale);
+ * theta = 0 (ll), 1 (lh - vertical), 2 (hh - diagonal), 3(hl - horizontal).
+ */
+av_always_inline static float dwt_quant_step(const struct dwt_model_params *params,
+                                             int lambda, int theta)
+{
+    /** Formula (1), page 1165 - display visual resolution (DVR),
+     * in pixels/degree of visual angle. This should be 56.55
+     */
+    float r = VIEW_DIST * REF_DISPLAY_HEIGHT * M_PI / 180.0;
+
+    /** Formula (9), page 1171 */
+    float temp = log10(pow(2.0, lambda+1) * params->f0 * params->g[theta] / r);
+    float Q = 2.0 * params->a * pow(10.0, params->k * temp * temp) /
+        dwt_7_9_basis_function_amplitudes[lambda][theta];
+
+    return Q;
+}
+
+static float get_cube(float val)
+{
+    return val * val * val;
+}
+
+static float adm_sum_cube(const float *x, int w, int h, int stride,
+                          double border_factor)
+{
+    int px_stride = stride / sizeof(float);
+    int left   = w * border_factor - 0.5;
+    int top    = h * border_factor - 0.5;
+    int right  = w - left;
+    int bottom = h - top;
+
+    int i, j;
+
+    float val;
+    float sum = 0;
+
+    for (i = top; i < bottom; i++) {
+        for (j = left; j < right; j++) {
+            val = fabsf(x[i * px_stride + j]);
+            sum += get_cube(val);
+        }
+    }
+
+    return powf(sum, 1.0f / 3.0f) + powf((bottom - top) * (right - left) /
+                                         32.0f, 1.0f / 3.0f);
+}
+
+static void adm_decouple(const adm_dwt_band_t *ref, const adm_dwt_band_t *main,
+                         const adm_dwt_band_t *r, const adm_dwt_band_t *a,
+                         int w, int h, int ref_stride, int main_stride,
+                         int r_stride, int a_stride)
+{
+    const float cos_1deg_sq = cos(1.0 * M_PI / 180.0) * cos(1.0 * M_PI / 180.0);
+    const float eps = 1e-30;
+
+    int ref_px_stride = ref_stride / sizeof(float);
+    int main_px_stride = main_stride / sizeof(float);
+    int r_px_stride = r_stride / sizeof(float);
+    int a_px_stride = a_stride / sizeof(float);
+
+    float oh, ov, od, th, tv, td;
+    float kh, kv, kd, tmph, tmpv, tmpd;
+    float ot_dp, o_mag_sq, t_mag_sq;
+    int angle_flag;
+    int i, j;
+
+    for (i = 0; i < h; i++) {
+        for (j = 0; j < w; j++) {
+            oh = ref->band_h[i * ref_px_stride + j];
+            ov = ref->band_v[i * ref_px_stride + j];
+            od = ref->band_d[i * ref_px_stride + j];
+            th = main->band_h[i * main_px_stride + j];
+            tv = main->band_v[i * main_px_stride + j];
+            td = main->band_d[i * main_px_stride + j];
+
+            kh = DIVS(th, oh + eps);
+            kv = DIVS(tv, ov + eps);
+            kd = DIVS(td, od + eps);
+
+            kh = kh < 0.0f ? 0.0f : (kh > 1.0f ? 1.0f : kh);
+
+
+            kv = kv < 0.0f ? 0.0f : (kv > 1.0f ? 1.0f : kv);
+            kd = kd < 0.0f ? 0.0f : (kd > 1.0f ? 1.0f : kd);
+
+            tmph = kh * oh;
+            tmpv = kv * ov;
+            tmpd = kd * od;
+            /** Determine if angle between (oh,ov) and (th,tv) is less than 1 degree.
+             * Given that u is the angle (oh,ov) and v is the angle (th,tv), this can
+             * be done by testing the inequvality.
+             *
+             * { (u.v.) >= 0 } AND { (u.v)^2 >= cos(1deg)^2 * ||u||^2 * ||v||^2 }
+             *
+             * Proof:
+             *
+             * cos(theta) = (u.v) / (||u|| * ||v||)
+             *
+             * IF u.v >= 0 THEN
+             *   cos(theta)^2 = (u.v)^2 / (||u||^2 * ||v||^2)
+             *   (u.v)^2 = cos(theta)^2 * ||u||^2 * ||v||^2
+             *
+             *   IF |theta| < 1deg THEN
+             *     (u.v)^2 >= cos(1deg)^2 * ||u||^2 * ||v||^2
+             *   END
+             * ELSE
+             *   |theta| > 90deg
+             * END
+             */
+            ot_dp = oh * th + ov * tv;
+            o_mag_sq = oh * oh + ov * ov;
+            t_mag_sq = th * th + tv * tv;
+
+            angle_flag = (ot_dp >= 0.0f) && (ot_dp * ot_dp >= cos_1deg_sq *
+                                             o_mag_sq * t_mag_sq);
+
+            if (angle_flag) {
+                tmph = th;
+                tmpv = tv;
+                tmpd = td;
+            }
+
+            r->band_h[i * r_px_stride + j] = tmph;
+            r->band_v[i * r_px_stride + j] = tmpv;
+            r->band_d[i * r_px_stride + j] = tmpd;
+
+            a->band_h[i * a_px_stride + j] = th - tmph;
+            a->band_v[i * a_px_stride + j] = tv - tmpv;
+            a->band_d[i * a_px_stride + j] = td - tmpd;
+        }
+    }
+}
+
+static void adm_csf(const adm_dwt_band_t *src, const adm_dwt_band_t *dst,
+                    int orig_h, int scale, int w, int h, int src_stride,
+                    int dst_stride)
+{
+    const float *src_angles[3] = { src->band_h, src->band_v, src->band_d };
+    float *dst_angles[3]       = { dst->band_h, dst->band_v, dst->band_d };
+
+    const float *src_ptr;
+    float *dst_ptr;
+
+    int src_px_stride = src_stride / sizeof(float);
+    int dst_px_stride = dst_stride / sizeof(float);
+
+    float factor1 = dwt_quant_step(&dwt_7_9_YCbCr_threshold[0], scale, 1);
+    float factor2 = dwt_quant_step(&dwt_7_9_YCbCr_threshold[0], scale, 2);
+    float rfactor[3] = {1.0f / factor1, 1.0f / factor1, 1.0f / factor2};
+
+    int i, j, theta;
+
+    for (theta = 0; theta < 3; theta++) {
+        src_ptr = src_angles[theta];
+        dst_ptr = dst_angles[theta];
+
+        for (i = 0; i < h; i++) {
+            for (j = 0; j < w; j++) {
+                dst_ptr[i * dst_px_stride + j] = rfactor[theta] *
+                    src_ptr[i * src_px_stride + j];
+            }
+        }
+    }
+}
+
+static void adm_cm_thresh(const adm_dwt_band_t *src, float *dst, int w, int h,
+                          int src_stride, int dst_stride)
+{
+    const float *angles[3] = { src->band_h, src->band_v, src->band_d };
+    const float *src_ptr;
+
+    int src_px_stride = src_stride / sizeof(float);
+    int dst_px_stride = dst_stride / sizeof(float);
+
+    float filt_coeff, img_coeff;
+
+    int theta, i, j, filt_i, filt_j, src_i, src_j;
+
+    for (i = 0; i < h; i++) {
+
+        for (j = 0; j < w; j++) {
+            dst[i * dst_px_stride + j] = 0;
+        }
+
+        for (theta = 0; theta < 3; ++theta) {
+            src_ptr = angles[theta];
+
+            for (j = 0; j < w; j++) {
+                float sum = 0;
+
+                for (filt_i = 0; filt_i < 3; filt_i++) {
+                    for (filt_j = 0; filt_j < 3; filt_j++) {
+                        filt_coeff = (filt_i == 1 && filt_j == 1) ? 1.0f / 15.0f : 1.0f /
+                            30.0f;
+
+                        src_i = i - 1 + filt_i;
+                        src_j = j - 1 + filt_j;
+
+                        src_i = FFABS(src_i);
+                        if (src_i >= h) {
+                            src_i = 2 * h - src_i - 1;
+                        }
+                        src_j = FFABS(src_j);
+                        if (src_j >= w) {
+                            src_j = 2 * w - src_j - 1;
+                        }
+                        img_coeff = fabsf(src_ptr[src_i * src_px_stride + src_j]);
+
+                        sum += filt_coeff * img_coeff;
+                    }
+                }
+
+                dst[i * dst_px_stride + j] += sum;
+            }
+        }
+    }
+}
+
+static void adm_cm(const adm_dwt_band_t *src, const adm_dwt_band_t *dst,
+                   const float *thresh, int w, int h, int src_stride,
+                   int dst_stride, int thresh_stride)
+{
+    int src_px_stride = src_stride / sizeof(float);
+    int dst_px_stride = dst_stride / sizeof(float);
+    int thresh_px_stride = thresh_stride / sizeof(float);
+
+    float xh, xv, xd, thr;
+
+    int i, j;
+
+    for (i = 0; i < h; i++) {
+        for (j = 0; j < w; j++) {
+            xh  = src->band_h[i * src_px_stride + j];
+            xv  = src->band_v[i * src_px_stride + j];
+            xd  = src->band_d[i * src_px_stride + j];
+            thr = thresh[i * thresh_px_stride + j];
+
+            xh = fabsf(xh) - thr;
+            xv = fabsf(xv) - thr;
+            xd = fabsf(xd) - thr;
+
+            xh = xh < 0.0f ? 0.0f : xh;
+            xv = xv < 0.0f ? 0.0f : xv;
+            xd = xd < 0.0f ? 0.0f : xd;
+
+            dst->band_h[i * dst_px_stride + j] = xh;
+            dst->band_v[i * dst_px_stride + j] = xv;
+            dst->band_d[i * dst_px_stride + j] = xd;
+        }
+    }
+}
+
+static void adm_dwt2(const float *src, const adm_dwt_band_t *dst, int w, int h,
+                     int src_stride, int dst_stride, float *temp_lo,
+                     float* temp_hi)
+{
+    const float *filter_lo = dwt2_db2_coeffs_lo;
+    const float *filter_hi = dwt2_db2_coeffs_hi;
+    int filt_w = sizeof(dwt2_db2_coeffs_lo) / sizeof(float);
+
+    int src_px_stride = src_stride / sizeof(float);
+    int dst_px_stride = dst_stride / sizeof(float);
+
+    float filt_coeff_lo, filt_coeff_hi, img_coeff;
+
+    int i, j, filt_i, filt_j, src_i, src_j;
+
+    for (i = 0; i < (h + 1) / 2; i++) {
+        /** Vertical pass. */
+        for (j = 0; j < w; j++) {
+            float sum_lo = 0;
+            float sum_hi = 0;
+
+            for (filt_i = 0; filt_i < filt_w; filt_i++) {
+                filt_coeff_lo = filter_lo[filt_i];
+                filt_coeff_hi = filter_hi[filt_i];
+
+                src_i = 2 * i - 1 + filt_i;
+
+                src_i = FFABS(src_i);
+                if (src_i >= h) {
+                    src_i = 2 * h - src_i - 1;
+                }
+
+                img_coeff = src[src_i * src_px_stride + j];
+
+                sum_lo += filt_coeff_lo * img_coeff;
+                sum_hi += filt_coeff_hi * img_coeff;
+            }
+
+            temp_lo[j] = sum_lo;
+            temp_hi[j] = sum_hi;
+        }
+
+        /** Horizontal pass (lo). */
+        for (j = 0; j < (w + 1) / 2; j++) {
+            float sum_lo = 0;
+            float sum_hi = 0;
+
+            for (filt_j = 0; filt_j < filt_w; filt_j++) {
+                filt_coeff_lo = filter_lo[filt_j];
+                filt_coeff_hi = filter_hi[filt_j];
+
+                src_j = 2 * j - 1 + filt_j;
+
+                src_j = FFABS(src_j);
+                if (src_j >= w) {
+                    src_j = 2 * w - src_j - 1;
+                }
+
+                img_coeff = temp_lo[src_j];
+
+                sum_lo += filt_coeff_lo * img_coeff;
+                sum_hi += filt_coeff_hi * img_coeff;
+            }
+
+            dst->band_a[i * dst_px_stride + j] = sum_lo;
+            dst->band_v[i * dst_px_stride + j] = sum_hi;
+        }
+
+        /** Horizontal pass (hi). */
+        for (j = 0; j < (w + 1) / 2; j++) {
+            float sum_lo = 0;
+            float sum_hi = 0;
+
+            for (filt_j = 0; filt_j < filt_w; filt_j++) {
+                filt_coeff_lo = filter_lo[filt_j];
+                filt_coeff_hi = filter_hi[filt_j];
+
+                src_j = 2 * j - 1 + filt_j;
+
+                src_j = FFABS(src_j);
+                if (src_j >= w) {
+                    src_j = 2 * w - src_j - 1;
+                }
+
+                img_coeff = temp_hi[src_j];
+
+                sum_lo += filt_coeff_lo * img_coeff;
+                sum_hi += filt_coeff_hi * img_coeff;
+            }
+
+            dst->band_h[i * dst_px_stride + j] = sum_lo;
+            dst->band_d[i * dst_px_stride + j] = sum_hi;
+        }
+    }
+
+}
+
+static void adm_buffer_copy(const void *src, void *dst, int linewidth, int h,
+                            int src_stride, int dst_stride)
+{
+    const char *src_p = src;
+    char *dst_p = dst;
+    int i;
+
+    for (i = 0; i < h; i++) {
+        memcpy(dst_p, src_p, linewidth);
+        src_p += src_stride;
+        dst_p += dst_stride;
+    }
+}
+
+static char *init_dwt_band(adm_dwt_band_t *band, char *data_top, size_t buf_sz)
+{
+    band->band_a = (float *) data_top;
+    data_top += buf_sz;
+    band->band_h = (float *) data_top;
+    data_top += buf_sz;
+    band->band_v = (float *) data_top;
+    data_top += buf_sz;
+    band->band_d = (float *) data_top;
+    data_top += buf_sz;
+    return data_top;
+}
+
+int compute_adm2(const float *ref, const float *main, int w, int h,
+                 int ref_stride, int main_stride, double *score,
+                 double *score_num, double *score_den, double *scores,
+                 float *data_buf, float *temp_lo, float* temp_hi)
+{
+    double numden_limit = 1e-2 * (w * h) / (1920.0 * 1080.0);
+
+    char *data_top;
+
+    float *ref_scale;
+    float *main_scale;
+
+    adm_dwt_band_t ref_dwt2;
+    adm_dwt_band_t main_dwt2;
+
+    adm_dwt_band_t decouple_r;
+    adm_dwt_band_t decouple_a;
+
+    adm_dwt_band_t csf_o;
+    adm_dwt_band_t csf_r;
+    adm_dwt_band_t csf_a;
+
+    float *mta;
+
+    adm_dwt_band_t cm_r;
+
+    const float *curr_ref_scale = (float *) ref;
+    const float *curr_main_scale = (float *) main;
+    int curr_ref_stride = ref_stride;
+    int curr_main_stride = main_stride;
+
+    int orig_h = h;
+
+    int buf_stride = ALIGN_CEIL(((w + 1) / 2) * sizeof(float));
+    size_t buf_sz = (size_t)buf_stride * ((h + 1) / 2);
+
+    double num = 0;
+    double den = 0;
+
+    int scale;
+    int ret = 1;
+
+    data_top = (char *) (data_buf);
+
+    ref_scale = (float *) data_top;
+    data_top += buf_sz;
+    main_scale = (float *) data_top;
+    data_top += buf_sz;
+
+    data_top = init_dwt_band(&ref_dwt2, data_top, buf_sz);
+    data_top = init_dwt_band(&main_dwt2, data_top, buf_sz);
+    data_top = init_dwt_band(&decouple_r, data_top, buf_sz);
+    data_top = init_dwt_band(&decouple_a, data_top, buf_sz);
+    data_top = init_dwt_band(&csf_o, data_top, buf_sz);
+    data_top = init_dwt_band(&csf_r, data_top, buf_sz);
+    data_top = init_dwt_band(&csf_a, data_top, buf_sz);
+
+    mta = (float *) data_top;
+    data_top += buf_sz;
+
+    data_top = init_dwt_band(&cm_r, data_top, buf_sz);
+
+    for (scale = 0; scale < 4; scale++) {
+        float num_scale = 0.0;
+        float den_scale = 0.0;
+
+        adm_dwt2(curr_ref_scale, &ref_dwt2, w, h, curr_ref_stride, buf_stride,
+                 temp_lo, temp_hi);
+        adm_dwt2(curr_main_scale, &main_dwt2, w, h, curr_main_stride, buf_stride,
+                 temp_lo, temp_hi);
+
+        w = (w + 1) / 2;
+        h = (h + 1) / 2;
+
+        adm_decouple(&ref_dwt2, &main_dwt2, &decouple_r, &decouple_a, w, h,
+                     buf_stride, buf_stride, buf_stride, buf_stride);
+
+        adm_csf(&ref_dwt2, &csf_o, orig_h, scale, w, h, buf_stride, buf_stride);
+        adm_csf(&decouple_r, &csf_r, orig_h, scale, w, h, buf_stride, buf_stride);
+        adm_csf(&decouple_a, &csf_a, orig_h, scale, w, h, buf_stride, buf_stride);
+
+        adm_cm_thresh(&csf_a, mta, w, h, buf_stride, buf_stride);
+        adm_cm(&csf_r, &cm_r, mta, w, h, buf_stride, buf_stride, buf_stride);
+
+        num_scale += adm_sum_cube(cm_r.band_h, w, h, buf_stride, ADM_BORDER_FACTOR);
+        num_scale += adm_sum_cube(cm_r.band_v, w, h, buf_stride, ADM_BORDER_FACTOR);
+        num_scale += adm_sum_cube(cm_r.band_d, w, h, buf_stride, ADM_BORDER_FACTOR);
+
+        den_scale += adm_sum_cube(csf_o.band_h, w, h, buf_stride, ADM_BORDER_FACTOR);
+        den_scale += adm_sum_cube(csf_o.band_v, w, h, buf_stride, ADM_BORDER_FACTOR);
+        den_scale += adm_sum_cube(csf_o.band_d, w, h, buf_stride, ADM_BORDER_FACTOR);
+
+        num += num_scale;
+        den += den_scale;
+
+        adm_buffer_copy(ref_dwt2.band_a, ref_scale, w * sizeof(float), h,
+                        buf_stride, buf_stride);
+        adm_buffer_copy(main_dwt2.band_a, main_scale, w * sizeof(float), h,
+                        buf_stride, buf_stride);
+
+        curr_ref_scale = ref_scale;
+        curr_main_scale = main_scale;
+        curr_ref_stride = buf_stride;
+        curr_main_stride = buf_stride;
+
+        scores[2*scale+0] = num_scale;
+        scores[2*scale+1] = den_scale;
+    }
+
+    num = num < numden_limit ? 0 : num;
+    den = den < numden_limit ? 0 : den;
+
+    if (den == 0.0) {
+        *score = 1.0f;
+    } else {
+        *score = num / den;
+    }
+    *score_num = num;
+    *score_den = den;
+
+    ret = 0;
+
+    return ret;
+}
+
+#define offset_fn(type, bits) \
+    static void offset_##bits##bit(ADMContext *s, const AVFrame *ref, AVFrame *main, int stride) \
+{ \
+    int w = s->width; \
+    int h = s->height; \
+    int i,j; \
+    \
+    int ref_stride = ref->linesize[0]; \
+    int main_stride = main->linesize[0]; \
+    \
+    const type *ref_ptr = (const type *) ref->data[0]; \
+    const type *main_ptr = (const type *) main->data[0]; \
+    \
+    float *ref_ptr_data = s->ref_data; \
+    float *main_ptr_data = s->main_data; \
+    \
+    for(i = 0; i < h; i++) { \
+        for(j = 0; j < w; j++) { \
+            ref_ptr_data[j] = (float) ref_ptr[j] + OPT_RANGE_PIXEL_OFFSET; \
+            main_ptr_data[j] = (float) main_ptr[j] + OPT_RANGE_PIXEL_OFFSET; \
+        } \
+        ref_ptr += ref_stride / sizeof(type); \
+        ref_ptr_data += stride / sizeof(float); \
+        main_ptr += main_stride / sizeof(type); \
+        main_ptr_data += stride / sizeof(float); \
+    } \
+}
+
+offset_fn(uint8_t, 8);
+offset_fn(uint16_t, 10);
+
+static void set_meta(AVDictionary **metadata, const char *key, float d)
+{
+    char value[128];
+    snprintf(value, sizeof(value), "%0.2f", d);
+    av_dict_set(metadata, key, value, 0);
+}
+
+static AVFrame *do_adm(AVFilterContext *ctx, AVFrame *main, const AVFrame *ref)
+{
+    ADMContext *s = ctx->priv;
+    AVDictionary **metadata = &main->metadata;
+
+    double score = 0.0;
+    double score_num = 0;
+    double score_den = 0;
+    double scores[2*4];
+
+    int w = s->width;
+    int h = s->height;
+
+    int stride;
+
+    stride = ALIGN_CEIL(w * sizeof(float));
+
+    /** Offset ref and main pixels by OPT_RANGE_PIXEL_OFFSET */
+    if (s->desc->comp[0].depth <= 8) {
+        offset_8bit(s, ref, main, stride);
+    } else {
+        offset_10bit(s, ref, main, stride);
+    }
+
+    compute_adm2(s->ref_data, s->main_data, w, h, stride, stride, &score,
+                 &score_num, &score_den, scores, s->data_buf, s->temp_lo,
+                 s->temp_hi);
+
+    set_meta(metadata, "lavfi.adm.score", score);
+
+    s->nb_frames++;
+
+    s->adm_sum += score;
+
+    return main;
+}
+
+static av_cold int init(AVFilterContext *ctx)
+{
+    ADMContext *s = ctx->priv;
+
+    s->dinput.process = do_adm;
+
+    return 0;
+}
+
+static int query_formats(AVFilterContext *ctx)
+{
+    static const enum AVPixelFormat pix_fmts[] = {
+        AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV420P,
+        AV_PIX_FMT_YUV444P10LE, AV_PIX_FMT_YUV422P10LE, AV_PIX_FMT_YUV420P10LE,
+        AV_PIX_FMT_NONE
+    };
+
+    AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts);
+    if (!fmts_list)
+        return AVERROR(ENOMEM);
+    return ff_set_common_formats(ctx, fmts_list);
+}
+
+static int config_input_ref(AVFilterLink *inlink)
+{
+    AVFilterContext *ctx  = inlink->dst;
+    ADMContext *s = ctx->priv;
+    int buf_stride;
+    size_t buf_sz;
+    int stride;
+    size_t data_sz;
+
+    if (ctx->inputs[0]->w != ctx->inputs[1]->w ||
+        ctx->inputs[0]->h != ctx->inputs[1]->h) {
+        av_log(ctx, AV_LOG_ERROR, "Width and height of input videos must be same.\n");
+        return AVERROR(EINVAL);
+    }
+    if (ctx->inputs[0]->format != ctx->inputs[1]->format) {
+        av_log(ctx, AV_LOG_ERROR, "Inputs must be of same pixel format.\n");
+        return AVERROR(EINVAL);
+    }
+
+    s->desc = av_pix_fmt_desc_get(inlink->format);
+    s->width = ctx->inputs[0]->w;
+    s->height = ctx->inputs[0]->h;
+
+    stride = ALIGN_CEIL(s->width * sizeof(float));
+    data_sz = (size_t)stride * s->height;
+
+    if (!(s->ref_data = av_malloc(data_sz))) {
+        av_log(ctx, AV_LOG_ERROR, "ref data allocation failed.\n");
+        return AVERROR(ENOMEM);
+    }
+
+    if (!(s->main_data = av_malloc(data_sz))) {
+        av_log(ctx, AV_LOG_ERROR, "main data allocation failed.\n");
+        return AVERROR(ENOMEM);
+    }
+
+    buf_stride = ALIGN_CEIL(((s->width + 1) / 2) * sizeof(float));
+    buf_sz = (size_t)buf_stride * ((s->height + 1) / 2);
+
+    if (SIZE_MAX / buf_sz < 35) {
+        av_log(ctx, AV_LOG_ERROR, "error: SIZE_MAX / buf_sz_one < 35, buf_sz_one = %lu.\n", buf_sz);
+        return AVERROR(EINVAL);
+    }
+
+    if (!(s->data_buf = av_malloc(buf_sz * 35))) {
+        av_log(ctx, AV_LOG_ERROR, "data_buf allocation failed.\n");
+        return AVERROR(ENOMEM);
+    }
+
+    if (!(s->temp_lo = av_malloc(stride))) {
+        av_log(ctx, AV_LOG_ERROR, "temp lo allocation failed.\n");
+        return AVERROR(ENOMEM);
+    }
+
+    if (!(s->temp_hi = av_malloc(stride))) {
+        av_log(ctx, AV_LOG_ERROR, "temp hi allocation failed.\n");
+        return AVERROR(ENOMEM);
+    }
+
+    return 0;
+}
+
+
+static int config_output(AVFilterLink *outlink)
+{
+    AVFilterContext *ctx = outlink->src;
+    ADMContext *s = ctx->priv;
+    AVFilterLink *mainlink = ctx->inputs[0];
+    int ret;
+
+    outlink->w = mainlink->w;
+    outlink->h = mainlink->h;
+    outlink->time_base = mainlink->time_base;
+    outlink->sample_aspect_ratio = mainlink->sample_aspect_ratio;
+    outlink->frame_rate = mainlink->frame_rate;
+    if ((ret = ff_dualinput_init(ctx, &s->dinput)) < 0)
+        return ret;
+
+    return 0;
+}
+
+static int filter_frame(AVFilterLink *inlink, AVFrame *inpicref)
+{
+    ADMContext *s = inlink->dst->priv;
+    return ff_dualinput_filter_frame(&s->dinput, inlink, inpicref);
+}
+
+static int request_frame(AVFilterLink *outlink)
+{
+    ADMContext *s = outlink->src->priv;
+    return ff_dualinput_request_frame(&s->dinput, outlink);
+}
+
+static av_cold void uninit(AVFilterContext *ctx)
+{
+    ADMContext *s = ctx->priv;
+
+    if (s->nb_frames > 0) {
+        av_log(ctx, AV_LOG_INFO, "ADM AVG: %.3f\n", s->adm_sum / s->nb_frames);
+    }
+
+    av_free(s->ref_data);
+    av_free(s->main_data);
+    av_free(s->data_buf);
+    av_free(s->temp_lo);
+    av_free(s->temp_hi);
+
+    ff_dualinput_uninit(&s->dinput);
+}
+
+static const AVFilterPad adm_inputs[] = {
+    {
+        .name         = "main",
+        .type         = AVMEDIA_TYPE_VIDEO,
+        .filter_frame = filter_frame,
+    },{
+        .name         = "reference",
+        .type         = AVMEDIA_TYPE_VIDEO,
+        .filter_frame = filter_frame,
+        .config_props = config_input_ref,
+    },
+    { NULL }
+};
+
+static const AVFilterPad adm_outputs[] = {
+    {
+        .name          = "default",
+        .type          = AVMEDIA_TYPE_VIDEO,
+        .config_props  = config_output,
+        .request_frame = request_frame,
+    },
+    { NULL }
+};
+
+AVFilter ff_vf_adm = {
+    .name          = "adm",
+    .description   = NULL_IF_CONFIG_SMALL("Calculate the ADM between two video streams."),
+    .init          = init,
+    .uninit        = uninit,
+    .query_formats = query_formats,
+    .priv_size     = sizeof(ADMContext),
+    .priv_class    = &adm_class,
+    .inputs        = adm_inputs,
+    .outputs       = adm_outputs,
+};