Adds a motion compensated temporal denoiser to the encoder.

Some refactoring in rdopt.c and pickinter.c.

Change-Id: I4f50020eb3313c37f4d441d708fedcaf219d3038

configure

@@ -47,6 +47,7 @@ Advanced options:
   ${toggle_small}                 favor smaller size over speed
   ${toggle_postproc_visualizer}   macro block / block level visualizers
   ${toggle_multi_res_encoding}    enable multiple-resolution encoding
+  ${toggle_temporal_denoising}    enable temporal denoising and disable the spatial denoiser
 
 Codecs:
   Codecs can be selectively enabled or disabled individually, or by family:
@@ -165,6 +166,7 @@ enable md5
 enable spatial_resampling
 enable multithread
 enable os_support
+enable temporal_denoising
 
 [ -d ${source_path}/../include ] && enable alt_tree_layout
 for d in vp8; do
@@ -263,6 +265,7 @@ CONFIG_LIST="
     os_support
     unit_tests
     multi_res_encoding
+    temporal_denoising
 "
 CMDLINE_SELECT="
     extra_warnings
@@ -307,6 +310,7 @@ CMDLINE_SELECT="
     postproc_visualizer
     unit_tests
     multi_res_encoding
+    temporal_denoising
 "
 
 process_cmdline() {

vp8/common/alloccommon.c

@@ -37,7 +37,6 @@ static void update_mode_info_border(MODE_INFO *mi, int rows, int cols)
 void vp8_de_alloc_frame_buffers(VP8_COMMON *oci)
 {
     int i;
-
     for (i = 0; i < NUM_YV12_BUFFERS; i++)
         vp8_yv12_de_alloc_frame_buffer(&oci->yv12_fb[i]);
 

vp8/common/blockd.h

@@ -215,6 +215,12 @@ typedef struct macroblockd
     MODE_INFO *mode_info_context;
     int mode_info_stride;
 
+#if CONFIG_TEMPORAL_DENOISING
+    MB_PREDICTION_MODE best_sse_inter_mode;
+    int_mv best_sse_mv;
+    unsigned char need_to_clamp_best_mvs;
+#endif
+
     FRAME_TYPE frame_type;
 
     int up_available;

vp8/common/postproc.c

@@ -362,6 +362,7 @@ void vp8_deblock(YV12_BUFFER_CONFIG         *source,
     vp8_post_proc_down_and_across(source->v_buffer, post->v_buffer, source->uv_stride, post->uv_stride, source->uv_height, source->uv_width, ppl);
 }
 
+#if !(CONFIG_TEMPORAL_DENOISING)
 void vp8_de_noise(YV12_BUFFER_CONFIG         *source,
                   YV12_BUFFER_CONFIG         *post,
                   int                         q,
@@ -398,6 +399,7 @@ void vp8_de_noise(YV12_BUFFER_CONFIG         *source,
         source->uv_width - 4, ppl);
 
 }
+#endif
 
 double vp8_gaussian(double sigma, double mu, double x)
 {

vp8/encoder/denoising.c

+/*
+ *  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "denoising.h"
+
+#include "vp8/common/reconinter.h"
+#include "vpx/vpx_integer.h"
+#include "vpx_mem/vpx_mem.h"
+#include "vpx_rtcd.h"
+
+const unsigned int NOISE_MOTION_THRESHOLD = 20*20;
+const unsigned int NOISE_DIFF2_THRESHOLD = 75;
+// SSE_DIFF_THRESHOLD is selected as ~95% confidence assuming var(noise) ~= 100.
+const unsigned int SSE_DIFF_THRESHOLD = 16*16*20;
+const unsigned int SSE_THRESHOLD = 16*16*40;
+
+static __inline uint8_t blend(uint8_t state, uint8_t sample, uint8_t factor_q8)
+{
+  return (uint8_t)(
+      (((uint16_t)factor_q8 * ((uint16_t)state) +  // Q8
+        (uint16_t)(256 - factor_q8) * ((uint16_t)sample)) + 128)  // Q8
+      >> 8);
+}
+
+static unsigned int denoiser_motion_compensate(YV12_BUFFER_CONFIG* src,
+                                               YV12_BUFFER_CONFIG* dst,
+                                               MACROBLOCK* x,
+                                               unsigned int best_sse,
+                                               unsigned int zero_mv_sse,
+                                               int recon_yoffset,
+                                               int recon_uvoffset)
+{
+  MACROBLOCKD filter_xd = x->e_mbd;
+  int mv_col;
+  int mv_row;
+  int sse_diff = zero_mv_sse - best_sse;
+  // Compensate the running average.
+  filter_xd.pre.y_buffer = src->y_buffer + recon_yoffset;
+  filter_xd.pre.u_buffer = src->u_buffer + recon_uvoffset;
+  filter_xd.pre.v_buffer = src->v_buffer + recon_uvoffset;
+  // Write the compensated running average to the destination buffer.
+  filter_xd.dst.y_buffer = dst->y_buffer + recon_yoffset;
+  filter_xd.dst.u_buffer = dst->u_buffer + recon_uvoffset;
+  filter_xd.dst.v_buffer = dst->v_buffer + recon_uvoffset;
+  // Use the best MV for the compensation.
+  filter_xd.mode_info_context->mbmi.ref_frame = LAST_FRAME;
+  filter_xd.mode_info_context->mbmi.mode = filter_xd.best_sse_inter_mode;
+  filter_xd.mode_info_context->mbmi.mv = filter_xd.best_sse_mv;
+  filter_xd.mode_info_context->mbmi.need_to_clamp_mvs =
+      filter_xd.need_to_clamp_best_mvs;
+  mv_col = filter_xd.best_sse_mv.as_mv.col;
+  mv_row = filter_xd.best_sse_mv.as_mv.row;
+  if (filter_xd.mode_info_context->mbmi.mode <= B_PRED ||
+      (mv_row*mv_row + mv_col*mv_col <= NOISE_MOTION_THRESHOLD &&
+       sse_diff < SSE_DIFF_THRESHOLD))
+  {
+    // Handle intra blocks as referring to last frame with zero motion and
+    // let the absolute pixel difference affect the filter factor.
+    // Also consider small amount of motion as being random walk due to noise,
+    // if it doesn't mean that we get a much bigger error.
+    // Note that any changes to the mode info only affects the denoising.
+    filter_xd.mode_info_context->mbmi.ref_frame = LAST_FRAME;
+    filter_xd.mode_info_context->mbmi.mode = ZEROMV;
+    filter_xd.mode_info_context->mbmi.mv.as_int = 0;
+    x->e_mbd.best_sse_inter_mode = ZEROMV;
+    x->e_mbd.best_sse_mv.as_int = 0;
+    best_sse = zero_mv_sse;
+  }
+  if (!x->skip)
+  {
+    vp8_build_inter_predictors_mb(&filter_xd);
+  }
+  else
+  {
+    vp8_build_inter16x16_predictors_mb(&filter_xd,
+                                       filter_xd.dst.y_buffer,
+                                       filter_xd.dst.u_buffer,
+                                       filter_xd.dst.v_buffer,
+                                       filter_xd.dst.y_stride,
+                                       filter_xd.dst.uv_stride);
+  }
+  return best_sse;
+}
+
+static void denoiser_filter(YV12_BUFFER_CONFIG* mc_running_avg,
+                            YV12_BUFFER_CONFIG* running_avg,
+                            MACROBLOCK* signal,
+                            unsigned int motion_magnitude2,
+                            int y_offset,
+                            int uv_offset)
+{
+  unsigned char* sig = signal->thismb;
+  int sig_stride = 16;
+  unsigned char* mc_running_avg_y = mc_running_avg->y_buffer + y_offset;
+  int mc_avg_y_stride = mc_running_avg->y_stride;
+  unsigned char* running_avg_y = running_avg->y_buffer + y_offset;
+  int avg_y_stride = running_avg->y_stride;
+  int r, c;
+  for (r = 0; r < 16; r++)
+  {
+    for (c = 0; c < 16; c++)
+    {
+      int diff;
+      int absdiff = 0;
+      unsigned int filter_coefficient;
+      absdiff = sig[c] - mc_running_avg_y[c];
+      absdiff = absdiff > 0 ? absdiff : -absdiff;
+      assert(absdiff >= 0 && absdiff < 256);
+      filter_coefficient = (255 << 8) / (256 + ((absdiff * 330) >> 3));
+      // Allow some additional filtering of static blocks, or blocks with very
+      // small motion vectors.
+      filter_coefficient += filter_coefficient / (3 + (motion_magnitude2 >> 3));
+      filter_coefficient = filter_coefficient > 255 ? 255 : filter_coefficient;
+
+      running_avg_y[c] = blend(mc_running_avg_y[c], sig[c], filter_coefficient);
+      diff = sig[c] - running_avg_y[c];
+
+      if (diff * diff < NOISE_DIFF2_THRESHOLD)
+      {
+        // Replace with mean to suppress the noise.
+        sig[c] = running_avg_y[c];
+      }
+      else
+      {
+        // Replace the filter state with the signal since the change in this
+        // pixel isn't classified as noise.
+        running_avg_y[c] = sig[c];
+      }
+    }
+    sig += sig_stride;
+    mc_running_avg_y += mc_avg_y_stride;
+    running_avg_y += avg_y_stride;
+  }
+}
+
+int vp8_denoiser_allocate(VP8_DENOISER *denoiser, int width, int height)
+{
+  assert(denoiser);
+  denoiser->yv12_running_avg.flags = 0;
+  if (vp8_yv12_alloc_frame_buffer(&(denoiser->yv12_running_avg), width,
+                                  height, VP8BORDERINPIXELS) < 0)
+  {
+      vp8_denoiser_free(denoiser);
+      return 1;
+  }
+  denoiser->yv12_mc_running_avg.flags = 0;
+  if (vp8_yv12_alloc_frame_buffer(&(denoiser->yv12_mc_running_avg), width,
+                                  height, VP8BORDERINPIXELS) < 0)
+  {
+      vp8_denoiser_free(denoiser);
+      return 1;
+  }
+  vpx_memset(denoiser->yv12_running_avg.buffer_alloc, 0,
+             denoiser->yv12_running_avg.frame_size);
+  vpx_memset(denoiser->yv12_mc_running_avg.buffer_alloc, 0,
+             denoiser->yv12_mc_running_avg.frame_size);
+  return 0;
+}
+
+void vp8_denoiser_free(VP8_DENOISER *denoiser)
+{
+  assert(denoiser);
+  vp8_yv12_de_alloc_frame_buffer(&denoiser->yv12_running_avg);
+  vp8_yv12_de_alloc_frame_buffer(&denoiser->yv12_mc_running_avg);
+}
+
+void vp8_denoiser_denoise_mb(VP8_DENOISER *denoiser,
+                             MACROBLOCK *x,
+                             unsigned int best_sse,
+                             unsigned int zero_mv_sse,
+                             int recon_yoffset,
+                             int recon_uvoffset) {
+  int mv_row;
+  int mv_col;
+  unsigned int motion_magnitude2;
+  // Motion compensate the running average.
+  best_sse = denoiser_motion_compensate(&denoiser->yv12_running_avg,
+                                        &denoiser->yv12_mc_running_avg,
+                                        x,
+                                        best_sse,
+                                        zero_mv_sse,
+                                        recon_yoffset,
+                                        recon_uvoffset);
+
+  mv_row = x->e_mbd.best_sse_mv.as_mv.row;
+  mv_col = x->e_mbd.best_sse_mv.as_mv.col;
+  motion_magnitude2 = mv_row*mv_row + mv_col*mv_col;
+  if (best_sse > SSE_THRESHOLD ||
+      motion_magnitude2 > 8 * NOISE_MOTION_THRESHOLD)
+  {
+    // No filtering of this block since it differs too much from the predictor,
+    // or the motion vector magnitude is considered too big.
+    vp8_copy_mem16x16(x->thismb, 16,
+                      denoiser->yv12_running_avg.y_buffer + recon_yoffset,
+                      denoiser->yv12_running_avg.y_stride);
+    return;
+  }
+  // Filter.
+  denoiser_filter(&denoiser->yv12_mc_running_avg,
+                  &denoiser->yv12_running_avg,
+                  x,
+                  motion_magnitude2,
+                  recon_yoffset,
+                  recon_uvoffset);
+}

vp8/encoder/denoising.h

+/*
+ *  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP8_ENCODER_DENOISING_H_
+#define VP8_ENCODER_DENOISING_H_
+
+#include "block.h"
+
+typedef struct vp8_denoiser
+{
+  YV12_BUFFER_CONFIG yv12_running_avg;
+  YV12_BUFFER_CONFIG yv12_mc_running_avg;
+} VP8_DENOISER;
+
+int vp8_denoiser_allocate(VP8_DENOISER *denoiser, int width, int height);
+
+void vp8_denoiser_free(VP8_DENOISER *denoiser);
+
+void vp8_denoiser_denoise_mb(VP8_DENOISER *denoiser,
+                             MACROBLOCK *x,
+                             unsigned int best_sse,
+                             unsigned int zero_mv_sse,
+                             int recon_yoffset,
+                             int recon_uvoffset);
+
+#endif  // VP8_ENCODER_DENOISING_H_

vp8/encoder/encodeframe.c

@@ -1179,6 +1179,13 @@ int vp8cx_encode_inter_macroblock
     else
         x->encode_breakout = cpi->oxcf.encode_breakout;
 
+#if CONFIG_TEMPORAL_DENOISING
+    // Reset the best sse mode/mv for each macroblock.
+    x->e_mbd.best_sse_inter_mode = 0;
+    x->e_mbd.best_sse_mv.as_int = 0;
+    x->e_mbd.need_to_clamp_best_mvs = 0;
+#endif
+
     if (cpi->sf.RD)
     {
         int zbin_mode_boost_enabled = cpi->zbin_mode_boost_enabled;

vp8/encoder/onyx_if.c

@@ -1680,6 +1680,17 @@ void vp8_change_config(VP8_COMP *cpi, VP8_CONFIG *oxcf)
     cpi->alt_ref_source = NULL;
     cpi->is_src_frame_alt_ref = 0;
 
+#if CONFIG_TEMPORAL_DENOISING
+    if (cpi->oxcf.noise_sensitivity)
+    {
+      if (!cpi->denoiser.yv12_mc_running_avg.buffer_alloc)
+      {
+        int width = (cpi->oxcf.Width + 15) & ~15;
+        int height = (cpi->oxcf.Height + 15) & ~15;
+        vp8_denoiser_allocate(&cpi->denoiser, width, height);
+      }
+    }
+#endif
 
 #if 0
     // Experimental RD Code
@@ -2314,6 +2325,9 @@ void vp8_remove_compressor(VP8_COMP **ptr)
     vp8cx_remove_encoder_threads(cpi);
 #endif
 
+#if CONFIG_TEMPORAL_DENOISING
+    vp8_denoiser_free(&cpi->denoiser);
+#endif
     dealloc_compressor_data(cpi);
     vpx_free(cpi->mb.ss);
     vpx_free(cpi->tok);
@@ -3133,7 +3147,12 @@ void vp8_loopfilter_frame(VP8_COMP *cpi, VP8_COMMON *cm)
     }
 
     vp8_yv12_extend_frame_borders_ptr(cm->frame_to_show);
-
+#if CONFIG_TEMPORAL_DENOISING
+    if (cpi->oxcf.noise_sensitivity)
+    {
+      vp8_yv12_extend_frame_borders(&cpi->denoiser.yv12_running_avg);
+    }
+#endif
 }
 
 static void encode_frame_to_data_rate
@@ -3589,7 +3608,7 @@ static void encode_frame_to_data_rate
 
 
     scale_and_extend_source(cpi->un_scaled_source, cpi);
-#if !(CONFIG_REALTIME_ONLY) && CONFIG_POSTPROC
+#if !(CONFIG_REALTIME_ONLY) && CONFIG_POSTPROC && !(CONFIG_TEMPORAL_DENOISING)
 
     if (cpi->oxcf.noise_sensitivity > 0)
     {

vp8/encoder/onyx_int.h

@@ -28,6 +28,9 @@
 #include "mcomp.h"
 #include "vp8/common/findnearmv.h"
 #include "lookahead.h"
+#if CONFIG_TEMPORAL_DENOISING
+#include "vp8/encoder/denoising.h"
+#endif
 
 //#define SPEEDSTATS 1
 #define MIN_GF_INTERVAL             4
@@ -661,6 +664,10 @@ typedef struct VP8_COMP
 
     int droppable;
 
+#if CONFIG_TEMPORAL_DENOISING
+    VP8_DENOISER denoiser;
+#endif
+
     // Coding layer state variables
     unsigned int current_layer;
     LAYER_CONTEXT layer_context[MAX_LAYERS];

vp8/encoder/pickinter.c

@@ -24,6 +24,9 @@
 #include "mcomp.h"
 #include "rdopt.h"
 #include "vpx_mem/vpx_mem.h"
+#if CONFIG_TEMPORAL_DENOISING
+#include "denoising.h"
+#endif
 
 extern int VP8_UVSSE(MACROBLOCK *x);
 
@@ -450,6 +453,48 @@ void get_lower_res_motion_info(VP8_COMP *cpi, MACROBLOCKD *xd, int *dissim,
 }
 #endif
 
+static void check_for_encode_breakout(unsigned int sse, MACROBLOCK* x)
+{
+    if (sse < x->encode_breakout)
+    {
+        // Check u and v to make sure skip is ok
+        int sse2 = 0;
+
+        sse2 = VP8_UVSSE(x);
+
+        if (sse2 * 2 < x->encode_breakout)
+            x->skip = 1;
+        else
+            x->skip = 0;
+    }
+}
+
+static int evaluate_inter_mode(unsigned int* sse, int rate2, int* distortion2, VP8_COMP *cpi, MACROBLOCK *x)
+{
+    MB_PREDICTION_MODE this_mode = x->e_mbd.mode_info_context->mbmi.mode;
+    int_mv mv = x->e_mbd.mode_info_context->mbmi.mv;
+    int this_rd;
+    /* Exit early and don't compute the distortion if this macroblock
+     * is marked inactive. */
+    if (cpi->active_map_enabled && x->active_ptr[0] == 0)
+    {
+        *sse = 0;
+        *distortion2 = 0;
+        x->skip = 1;
+        return INT_MAX;
+    }
+
+    if((this_mode != NEWMV) ||
+        !(cpi->sf.half_pixel_search) || cpi->common.full_pixel==1)
+        *distortion2 = get_inter_mbpred_error(x,
+                                              &cpi->fn_ptr[BLOCK_16X16],
+                                              sse, mv);
+
+    this_rd = RDCOST(x->rdmult, x->rddiv, rate2, *distortion2);
+
+    check_for_encode_breakout(*sse, x);
+    return this_rd;
+}
 
 void vp8_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset,
                          int recon_uvoffset, int *returnrate,
@@ -476,7 +521,10 @@ void vp8_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset,
     int distortion2;
     int bestsme = INT_MAX;
     int best_mode_index = 0;
-    unsigned int sse = INT_MAX, best_sse = INT_MAX;
+    unsigned int sse = INT_MAX, best_rd_sse = INT_MAX;
+#if CONFIG_TEMPORAL_DENOISING
+    unsigned int zero_mv_sse = 0, best_sse = INT_MAX;
+#endif
 
     int_mv mvp;
 
@@ -488,9 +536,6 @@ void vp8_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset,
     int ref_frame_map[4];
     int sign_bias = 0;
 
-    int have_subp_search = cpi->sf.half_pixel_search;  /* In real-time mode,
-                                       when Speed >= 15, no sub-pixel search. */
-
 #if CONFIG_MULTI_RES_ENCODING
     int dissim = INT_MAX;
     int parent_ref_frame = 0;
@@ -657,7 +702,7 @@ void vp8_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset,
         {
         case B_PRED:
             /* Pass best so far to pick_intra4x4mby_modes to use as breakout */
-            distortion2 = best_sse;
+            distortion2 = best_rd_sse;
             pick_intra4x4mby_modes(x, &rate, &distortion2);
 
             if (distortion2 == INT_MAX)
@@ -905,43 +950,38 @@ void vp8_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset,
             rate2 += vp8_cost_mv_ref(this_mode, mdcounts);
             x->e_mbd.mode_info_context->mbmi.mv.as_int =
                                                     mode_mv[this_mode].as_int;
-
-            /* Exit early and don't compute the distortion if this macroblock
-             * is marked inactive. */
-            if (cpi->active_map_enabled && x->active_ptr[0] == 0)
-            {
-                sse = 0;
-                distortion2 = 0;
-                x->skip = 1;
-                break;
-            }
-
-            if((this_mode != NEWMV) ||
-                !(have_subp_search) || cpi->common.full_pixel==1)
-                distortion2 = get_inter_mbpred_error(x,
-                                                     &cpi->fn_ptr[BLOCK_16X16],
-                                                     &sse, mode_mv[this_mode]);
-
-            this_rd = RDCOST(x->rdmult, x->rddiv, rate2, distortion2);
-
-            if (sse < x->encode_breakout)
-            {
-                // Check u and v to make sure skip is ok
-                int sse2 = 0;
-
-                sse2 = VP8_UVSSE(x);
-
-                if (sse2 * 2 < x->encode_breakout)
-                    x->skip = 1;
-                else
-                    x->skip = 0;
-            }
+            this_rd = evaluate_inter_mode(&sse, rate2, &distortion2, cpi, x);
 
             break;
         default:
             break;
         }
 
+#if CONFIG_TEMPORAL_DENOISING
+        if (cpi->oxcf.noise_sensitivity)
+        {
+          // Store for later use by denoiser.
+          if (this_mode == ZEROMV &&
+              x->e_mbd.mode_info_context->mbmi.ref_frame == LAST_FRAME)
+          {
+            zero_mv_sse = sse;
+          }
+
+          // Store the best NEWMV in x for later use in the denoiser.
+          // We are restricted to the LAST_FRAME since the denoiser only keeps
+          // one filter state.
+          if (x->e_mbd.mode_info_context->mbmi.mode == NEWMV &&
+              x->e_mbd.mode_info_context->mbmi.ref_frame == LAST_FRAME)
+          {
+            best_sse = sse;
+            x->e_mbd.best_sse_inter_mode = NEWMV;
+            x->e_mbd.best_sse_mv = x->e_mbd.mode_info_context->mbmi.mv;
+            x->e_mbd.need_to_clamp_best_mvs =
+                x->e_mbd.mode_info_context->mbmi.need_to_clamp_mvs;
+          }
+        }
+#endif
+
         if (this_rd < best_rd || x->skip)
         {
             // Note index of best mode
@@ -949,7 +989,7 @@ void vp8_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset,
 
             *returnrate = rate2;
             *returndistortion = distortion2;
-            best_sse = sse;
+            best_rd_sse = sse;
             best_rd = this_rd;
             vpx_memcpy(&best_mbmode, &x->e_mbd.mode_info_context->mbmi,
                        sizeof(MB_MODE_INFO));
@@ -1011,6 +1051,43 @@ void vp8_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset,
         cpi->error_bins[this_rdbin] ++;
     }
 
+#if CONFIG_TEMPORAL_DENOISING
+    if (cpi->oxcf.noise_sensitivity)
+    {
+      if (x->e_mbd.best_sse_inter_mode == DC_PRED) {
+        // No best MV found.
+        x->e_mbd.best_sse_inter_mode = best_mbmode.mode;
+        x->e_mbd.best_sse_mv = best_mbmode.mv;
+        x->e_mbd.need_to_clamp_best_mvs = best_mbmode.need_to_clamp_mvs;
+        best_sse = best_rd_sse;
+      }
+      vp8_denoiser_denoise_mb(&cpi->denoiser, x, best_sse, zero_mv_sse,
+                              recon_yoffset, recon_uvoffset);
+
+      // Reevaluate ZEROMV after denoising.
+      if (best_mbmode.ref_frame == INTRA_FRAME)
+      {
+        int this_rd = 0;
+        rate2 = 0;
+        distortion2 = 0;
+        x->e_mbd.mode_info_context->mbmi.ref_frame = LAST_FRAME;
+        rate2 += x->ref_frame_cost[x->e_mbd.mode_info_context->mbmi.ref_frame];
+        this_mode = ZEROMV;
+        rate2 += vp8_cost_mv_ref(this_mode, mdcounts);
+        x->e_mbd.mode_info_context->mbmi.mode = this_mode;
+        x->e_mbd.mode_info_context->mbmi.uv_mode = DC_PRED;
+        x->e_mbd.mode_info_context->mbmi.mv.as_int = 0;
+        this_rd = evaluate_inter_mode(&sse, rate2, &distortion2, cpi, x);
+
+        if (this_rd < best_rd || x->skip)
+        {
+            vpx_memcpy(&best_mbmode, &x->e_mbd.mode_info_context->mbmi,
+                       sizeof(MB_MODE_INFO));
+        }
+      }
+    }
+#endif
+
     if (cpi->is_src_frame_alt_ref &&
         (best_mbmode.mode != ZEROMV || best_mbmode.ref_frame != ALTREF_FRAME))
     {

vp8/vp8_cx_iface.c

@@ -183,13 +183,19 @@ static vpx_codec_err_t validate_config(vpx_codec_alg_priv_t      *ctx,