ref: 6b33ca395fadf30b83baf6dd006a771c78ae202c
parent: 9aa58f3fcb2757870eebc2ea1a02b958064bc0e8
author: Deb Mukherjee <debargha@google.com>
date: Tue Apr 3 10:02:58 EDT 2012
Fixes to disable MFQE when there is motion. This patch includes: 1. fixes to disable block based termporal mixing when motion is detected (because this version of mfqe only handles zero motion). 2. The criterion used for determining whether to mix or not are changed to use squared differences rather than absolute differences. 3. Additional checks on color mismatch and excessive block flatness added. If the block as decoded has very low activity it is unlikely to yield benefits for mixing. Change-Id: I07331e5ab5ba64844c56e84b1a4b7de823eac6cb
--- a/vp8/common/mfqe.c
+++ b/vp8/common/mfqe.c
@@ -26,7 +26,6 @@
#include <limits.h>
#include <stdlib.h>
-
static void filter_by_weight(unsigned char *src, int src_stride,
unsigned char *dst, int dst_stride,
int block_size, int src_weight)
@@ -96,6 +95,27 @@
}
}
+static unsigned int int_sqrt(unsigned int x)
+{+ unsigned int y = x;
+ unsigned int guess;
+ int p = 1;
+ while (y>>=1) p++;
+ p>>=1;
+
+ guess=0;
+ while (p>=0)
+ {+ guess |= (1<<p);
+ if (x<guess*guess)
+ guess -= (1<<p);
+ p--;
+ }
+ /* choose between guess or guess+1 */
+ return guess+(guess*guess+guess+1<=x);
+}
+
+#define USE_SSD
static void multiframe_quality_enhance_block
(
int blksize, /* Currently only values supported are 16, 8, 4 */
@@ -117,7 +137,7 @@
{0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
};
- int blksizeby2 = blksize >> 1;
+ int uvblksize = blksize >> 1;
int qdiff = qcurr - qprev;
int i, j;
@@ -128,178 +148,128 @@
unsigned char *vp;
unsigned char *vdp;
- unsigned int act_sum = 0, sse, sad_sum = 0, thr, uvsad = UINT_MAX;
- unsigned int act[4], sad[4];
+ unsigned int act, actd, sad, usad, vsad, sse, thr, thrsq, actrisk;
if (blksize == 16)
{- act[0] = (vp8_variance8x8(yd, yd_stride,
- VP8_ZEROS, 0, &sse) + 32) >> 6;
- act_sum = act[0];
- sad[0] = (vp8_sad8x8(y, y_stride,
- yd, yd_stride, INT_MAX) + 32) >> 6;
- sad_sum = sad[0];
-
- act[1] = (vp8_variance8x8(yd + 8, yd_stride,
- VP8_ZEROS, 0, &sse) + 32) >> 6;
- act_sum += act[1];
- sad[1] = (vp8_sad8x8(y + 8, y_stride,
- yd + 8, yd_stride, INT_MAX) + 32) >> 6;
- sad_sum = sad[1];
-
- act[2] = (vp8_variance8x8(yd + 8 * yd_stride, yd_stride,
- VP8_ZEROS, 0, &sse) + 32) >> 6;
- act_sum += act[2];
- sad[2] = (vp8_sad8x8(y + 8 * y_stride, y_stride,
- yd + 8 * yd_stride, yd_stride, INT_MAX) + 32) >> 6;
- sad_sum = sad[2];
-
- act[3] = (vp8_variance8x8(yd + 8 * yd_stride + 8, yd_stride,
- VP8_ZEROS, 0, &sse) + 32) >> 6;
- act_sum += act[3];
- sad[3] = (vp8_sad8x8(y + 8 * y_stride + 8, y_stride,
- yd + 8 * y_stride + 8, yd_stride, INT_MAX)
- + 32) >> 6;
- sad_sum = sad[3];
+ actd = (vp8_variance16x16(yd, yd_stride, VP8_ZEROS, 0, &sse)+128)>>8;
+ act = (vp8_variance16x16(y, y_stride, VP8_ZEROS, 0, &sse)+128)>>8;
+#ifdef USE_SSD
+ sad = (vp8_variance16x16(y, y_stride, yd, yd_stride, &sse));
+ sad = (sse + 128)>>8;
+ usad = (vp8_variance8x8(u, uv_stride, ud, uvd_stride, &sse));
+ usad = (sse + 32)>>6;
+ vsad = (vp8_variance8x8(v, uv_stride, vd, uvd_stride, &sse));
+ vsad = (sse + 32)>>6;
+#else
+ sad = (vp8_sad16x16(y, y_stride, yd, yd_stride, INT_MAX)+128)>>8;
+ usad = (vp8_sad8x8(u, uv_stride, ud, uvd_stride, INT_MAX)+32)>>6;
+ vsad = (vp8_sad8x8(v, uv_stride, vd, uvd_stride, INT_MAX)+32)>>6;
+#endif
}
else if (blksize == 8)
{- act[0] = (vp8_variance4x4(yd, yd_stride,
- VP8_ZEROS, 0, &sse) + 4) >> 4;
- act_sum = act[0];
- sad[0] = (vp8_sad4x4(y, y_stride,
- yd, yd_stride, INT_MAX) + 4) >> 4;
- sad_sum = sad[0];
-
- act[1] = (vp8_variance4x4(yd + 4, yd_stride,
- VP8_ZEROS, 0, &sse) + 4) >> 4;
- act_sum += act[1];
- sad[1] = (vp8_sad4x4(y + 4, y_stride,
- yd + 4, yd_stride, INT_MAX) + 4) >> 4;
- sad_sum = sad[1];
-
- act[2] = (vp8_variance4x4(yd + 4 * yd_stride, yd_stride,
- VP8_ZEROS, 0, &sse) + 4) >> 4;
- act_sum += act[2];
- sad[2] = (vp8_sad4x4(y + 4 * y_stride, y_stride,
- yd + 4 * yd_stride, yd_stride, INT_MAX) + 4) >> 4;
- sad_sum = sad[2];
-
- act[3] = (vp8_variance4x4(yd + 4 * yd_stride + 4, yd_stride,
- VP8_ZEROS, 0, &sse) + 4) >> 4;
- act_sum += act[3];
- sad[3] = (vp8_sad4x4(y + 4 * y_stride + 4, y_stride,
- yd + 4 * y_stride + 4, yd_stride, INT_MAX)
- + 4) >> 4;
- sad_sum = sad[3];
+ actd = (vp8_variance8x8(yd, yd_stride, VP8_ZEROS, 0, &sse)+32)>>6;
+ act = (vp8_variance8x8(y, y_stride, VP8_ZEROS, 0, &sse)+32)>>6;
+#ifdef USE_SSD
+ sad = (vp8_variance8x8(y, y_stride, yd, yd_stride, &sse));
+ sad = (sse + 32)>>6;
+ usad = (vp8_variance4x4(u, uv_stride, ud, uvd_stride, &sse));
+ usad = (sse + 8)>>4;
+ vsad = (vp8_variance4x4(v, uv_stride, vd, uvd_stride, &sse));
+ vsad = (sse + 8)>>4;
+#else
+ sad = (vp8_sad8x8(y, y_stride, yd, yd_stride, INT_MAX)+32)>>6;
+ usad = (vp8_sad4x4(u, uv_stride, ud, uvd_stride, INT_MAX)+8)>>4;
+ vsad = (vp8_sad4x4(v, uv_stride, vd, uvd_stride, INT_MAX)+8)>>4;
+#endif
}
- else
- {- act_sum = (vp8_variance4x4(yd, yd_stride, VP8_ZEROS, 0, &sse) + 8) >> 4;
- sad_sum = (vp8_sad4x4(y, y_stride, yd, yd_stride, INT_MAX) + 8) >> 4;
- }
+ actrisk = (actd > act * 5);
+
/* thr = qdiff/8 + log2(act) + log4(qprev) */
thr = (qdiff >> 3);
-
+ while (actd >>= 1) thr++;
while (qprev >>= 2) thr++;
- if (blksize > 4)
+#ifdef USE_SSD
+ thrsq = thr * thr;
+ if (sad < thrsq &&
+ /* additional checks for color mismatch and excessive addition of
+ * high-frequencies */
+ 4 * usad < thrsq && 4 * vsad < thrsq && !actrisk)
+#else
+ if (sad < thr &&
+ /* additional checks for color mismatch and excessive addition of
+ * high-frequencies */
+ 2 * usad < thr && 2 * vsad < thr && !actrisk)
+#endif
{- unsigned int base_thr = thr, this_thr, this_act;
- int i;
+ int ifactor;
+#ifdef USE_SSD
+ /* TODO: optimize this later to not need sqr root */
+ sad = int_sqrt(sad);
+#endif
+ ifactor = (sad << MFQE_PRECISION) / thr;
+ ifactor >>= (qdiff >> 5);
- for (i = 0; i < 4; i++)
+ if (ifactor)
{- this_thr = base_thr;
- this_act = act[i];
-
- while (this_act >>= 1) this_thr++;
-
- if (sad[i] >= this_thr || act[i] < 16)
- {- sad_sum = UINT_MAX;
- break;
- }
+ apply_ifactor(y, y_stride, yd, yd_stride,
+ u, v, uv_stride,
+ ud, vd, uvd_stride,
+ blksize, ifactor);
}
}
-
- while (act_sum >>= 1) thr++;
-
- if (sad_sum < thr)
+ else /* else implicitly copy from previous frame */
{if (blksize == 16)
{- uvsad = (vp8_sad8x8(u, uv_stride, ud, uvd_stride, INT_MAX) + 32)
- >> 6;
-
- if (uvsad < thr)
- uvsad = (vp8_sad8x8(v, uv_stride, vd, uvd_stride, INT_MAX) + 32)
- >> 6;
+ vp8_copy_mem16x16(y, y_stride, yd, yd_stride);
+ vp8_copy_mem8x8(u, uv_stride, ud, uvd_stride);
+ vp8_copy_mem8x8(v, uv_stride, vd, uvd_stride);
}
- else
+ else /* if (blksize == 8) */
{- uvsad = (vp8_sad4x4(u, uv_stride, ud, uvd_stride, INT_MAX) + 8)
- >> 4;
-
- if (uvsad < thr)
- uvsad = (vp8_sad4x4(v, uv_stride, vd, uvd_stride, INT_MAX) + 8)
- >> 4;
+ vp8_copy_mem8x8(y, y_stride, yd, yd_stride);
+ for (up = u, udp = ud, i = 0; i < uvblksize; ++i, up += uv_stride, udp += uvd_stride)
+ vpx_memcpy(udp, up, uvblksize);
+ for (vp = v, vdp = vd, i = 0; i < uvblksize; ++i, vp += uv_stride, vdp += uvd_stride)
+ vpx_memcpy(vdp, vp, uvblksize);
}
}
+}
- if (uvsad < thr)
+static int qualify_inter_mb(const MODE_INFO *mode_info_context, int *map)
+{+ if (mode_info_context->mbmi.mb_skip_coeff)
+ map[0] = map[1] = map[2] = map[3] = 1;
+ else if (mode_info_context->mbmi.mode==SPLITMV)
{- static const int roundoff = (1 << (MFQE_PRECISION - 1));
- int ifactor = (sad_sum << MFQE_PRECISION) / thr;
- ifactor >>= (qdiff >> 5);
- // TODO: SIMD optimize this section
- if (ifactor)
+ static int ndx[4][4] =
{- int icfactor = (1 << MFQE_PRECISION) - ifactor;
- for (yp = y, ydp = yd, i = 0; i < blksize; ++i, yp += y_stride, ydp += yd_stride)
- {- for (j = 0; j < blksize; ++j)
- ydp[j] = (int)((yp[j] * ifactor + ydp[j] * icfactor + roundoff) >> MFQE_PRECISION);
- }
- for (up = u, udp = ud, i = 0; i < blksizeby2; ++i, up += uv_stride, udp += uvd_stride)
- {- for (j = 0; j < blksizeby2; ++j)
- udp[j] = (int)((up[j] * ifactor + udp[j] * icfactor + roundoff) >> MFQE_PRECISION);
- }
- for (vp = v, vdp = vd, i = 0; i < blksizeby2; ++i, vp += uv_stride, vdp += uvd_stride)
- {- for (j = 0; j < blksizeby2; ++j)
- vdp[j] = (int)((vp[j] * ifactor + vdp[j] * icfactor + roundoff) >> MFQE_PRECISION);
- }
+ {0, 1, 4, 5},+ {2, 3, 6, 7},+ {8, 9, 12, 13},+ {10, 11, 14, 15}+ };
+ int i, j;
+ for (i=0; i<4; ++i)
+ {+ map[i] = 1;
+ for (j=0; j<4 && map[j]; ++j)
+ map[i] &= (mode_info_context->bmi[ndx[i][j]].mv.as_mv.row <= 2 &&
+ mode_info_context->bmi[ndx[i][j]].mv.as_mv.col <= 2);
}
}
else
{- if (blksize == 16)
- {- vp8_copy_mem16x16(y, y_stride, yd, yd_stride);
- vp8_copy_mem8x8(u, uv_stride, ud, uvd_stride);
- vp8_copy_mem8x8(v, uv_stride, vd, uvd_stride);
- }
- else if (blksize == 8)
- {- vp8_copy_mem8x8(y, y_stride, yd, yd_stride);
- for (up = u, udp = ud, i = 0; i < blksizeby2; ++i, up += uv_stride, udp += uvd_stride)
- vpx_memcpy(udp, up, blksizeby2);
- for (vp = v, vdp = vd, i = 0; i < blksizeby2; ++i, vp += uv_stride, vdp += uvd_stride)
- vpx_memcpy(vdp, vp, blksizeby2);
- }
- else
- {- for (yp = y, ydp = yd, i = 0; i < blksize; ++i, yp += y_stride, ydp += yd_stride)
- vpx_memcpy(ydp, yp, blksize);
- for (up = u, udp = ud, i = 0; i < blksizeby2; ++i, up += uv_stride, udp += uvd_stride)
- vpx_memcpy(udp, up, blksizeby2);
- for (vp = v, vdp = vd, i = 0; i < blksizeby2; ++i, vp += uv_stride, vdp += uvd_stride)
- vpx_memcpy(vdp, vp, blksizeby2);
- }
+ map[0] = map[1] = map[2] = map[3] =
+ (mode_info_context->mbmi.mode > B_PRED &&
+ abs(mode_info_context->mbmi.mv.as_mv.row) <= 2 &&
+ abs(mode_info_context->mbmi.mv.as_mv.col) <= 2);
}
+ return (map[0]+map[1]+map[2]+map[3]);
}
void vp8_multiframe_quality_enhance
@@ -315,6 +285,7 @@
const MODE_INFO *mode_info_context = cm->mi;
int mb_row;
int mb_col;
+ int totmap, map[4];
int qcurr = cm->base_qindex;
int qprev = cm->postproc_state.last_base_qindex;
@@ -335,29 +306,50 @@
for (mb_col = 0; mb_col < cm->mb_cols; mb_col++)
{/* if motion is high there will likely be no benefit */
- if (((frame_type == INTER_FRAME &&
- abs(mode_info_context->mbmi.mv.as_mv.row) <= 10 &&
- abs(mode_info_context->mbmi.mv.as_mv.col) <= 10) ||
- (frame_type == KEY_FRAME)))
+ if (frame_type == INTER_FRAME) totmap = qualify_inter_mb(mode_info_context, map);
+ else totmap = (frame_type == KEY_FRAME ? 4 : 0);
+ if (totmap)
{- if (mode_info_context->mbmi.mode == B_PRED || mode_info_context->mbmi.mode == SPLITMV)
+ if (totmap < 4)
{int i, j;
for (i=0; i<2; ++i)
for (j=0; j<2; ++j)
- multiframe_quality_enhance_block(8, qcurr, qprev,
- y_ptr + 8*(i*show->y_stride+j),
- u_ptr + 4*(i*show->uv_stride+j),
- v_ptr + 4*(i*show->uv_stride+j),
- show->y_stride,
- show->uv_stride,
- yd_ptr + 8*(i*dest->y_stride+j),
- ud_ptr + 4*(i*dest->uv_stride+j),
- vd_ptr + 4*(i*dest->uv_stride+j),
- dest->y_stride,
- dest->uv_stride);
- }
- else
+ {+ if (map[i*2+j])
+ {+ multiframe_quality_enhance_block(8, qcurr, qprev,
+ y_ptr + 8*(i*show->y_stride+j),
+ u_ptr + 4*(i*show->uv_stride+j),
+ v_ptr + 4*(i*show->uv_stride+j),
+ show->y_stride,
+ show->uv_stride,
+ yd_ptr + 8*(i*dest->y_stride+j),
+ ud_ptr + 4*(i*dest->uv_stride+j),
+ vd_ptr + 4*(i*dest->uv_stride+j),
+ dest->y_stride,
+ dest->uv_stride);
+ }
+ else
+ {+ /* copy a 8x8 block */
+ int k;
+ unsigned char *up = u_ptr + 4*(i*show->uv_stride+j);
+ unsigned char *udp = ud_ptr + 4*(i*dest->uv_stride+j);
+ unsigned char *vp = v_ptr + 4*(i*show->uv_stride+j);
+ unsigned char *vdp = vd_ptr + 4*(i*dest->uv_stride+j);
+ vp8_copy_mem8x8(y_ptr + 8*(i*show->y_stride+j), show->y_stride,
+ yd_ptr + 8*(i*dest->y_stride+j), dest->y_stride);
+ for (k = 0; k < 4; ++k, up += show->uv_stride, udp += dest->uv_stride,
+ vp += show->uv_stride, vdp += dest->uv_stride)
+ {+ vpx_memcpy(udp, up, 4);
+ vpx_memcpy(vdp, vp, 4);
+ }
+ }
+ }
+ }
+ else /* totmap = 4 */
{multiframe_quality_enhance_block(16, qcurr, qprev, y_ptr,
u_ptr, v_ptr,
--- a/vp8/decoder/decodemv.c
+++ b/vp8/decoder/decodemv.c
@@ -660,4 +660,3 @@
mi++; /* skip left predictor each row */
}
}
-