Harmonize the global parameter mappings

lowres: -0.401% BDRATE (tempete -5.4%)

Also includes some fixes and cleanups.

Change-Id: I82922a453fad59fad2f12829d2ef1e56b2d20949

av1/common/mv.h

@@ -37,7 +37,7 @@ typedef struct mv32 {
 #if CONFIG_GLOBAL_MOTION || CONFIG_WARPED_MOTION
 // Bits of precision used for the model
 #define WARPEDMODEL_PREC_BITS 12
-#define WARPEDMODEL_ROW3HOMO_PREC_BITS 12
+#define WARPEDMODEL_ROW3HOMO_PREC_BITS 14
 
 // Bits of subpel precision for warped interpolation
 #define WARPEDPIXEL_PREC_BITS 6
@@ -57,12 +57,18 @@ typedef enum {
   AFFINE,       // affine, 6-parameter
   ROTZOOM,      // simplified affine with rotation and zoom only, 4-parameter
   TRANSLATION,  // translational motion 2-parameter
+  IDENTITY,     // identity transformation, 0-parameter
   TRANS_TYPES
 } TransformationType;
 
 // number of parameters used by each transformation in TransformationTypes
-static const int n_trans_model_params[TRANS_TYPES] = { 9, 6, 4, 2 };
+static const int n_trans_model_params[TRANS_TYPES] = { 8, 6, 4, 2, 0 };
 
+// The order of values in the wmmat matrix below is best described
+// by the homography:
+// [x'     (m2 m3 m0   [x
+//  y'  =   m4 m5 m1 =  y
+//  1]      m6 m7 1)    1]
 typedef struct {
   TransformationType wmtype;
   int32_t wmmat[8];  // For homography wmmat[9] is assumed to be 1
@@ -102,19 +108,29 @@ typedef struct {
 #define GM_ALPHA_PREC_DIFF (WARPEDMODEL_PREC_BITS - GM_ALPHA_PREC_BITS)
 #define GM_ALPHA_DECODE_FACTOR (1 << GM_ALPHA_PREC_DIFF)
 
-#define GM_ABS_ALPHA_BITS 9
+#define GM_ROW3HOMO_PREC_BITS 14
+#define GM_ROW3HOMO_PREC_DIFF \
+  (WARPEDMODEL_ROW3HOMO_PREC_BITS - GM_ROW3HOMO_PREC_BITS)
+#define GM_ROW3HOMO_DECODE_FACTOR (1 << GM_ROW3HOMO_PREC_DIFF)
+
 #define GM_ABS_TRANS_BITS 9
+#define GM_ABS_ALPHA_BITS 9
+#define GM_ABS_ROW3HOMO_BITS 9
 
 #define GM_TRANS_MAX (1 << GM_ABS_TRANS_BITS)
 #define GM_ALPHA_MAX (1 << GM_ABS_ALPHA_BITS)
+#define GM_ROW3HOMO_MAX (1 << GM_ABS_ROW3HOMO_BITS)
+
 #define GM_TRANS_MIN -GM_TRANS_MAX
 #define GM_ALPHA_MIN -GM_ALPHA_MAX
+#define GM_ROW3HOMO_MIN -GM_ROW3HOMO_MAX
 
 typedef enum {
   GLOBAL_ZERO = 0,
   GLOBAL_TRANSLATION = 1,
   GLOBAL_ROTZOOM = 2,
   GLOBAL_AFFINE = 3,
+  // GLOBAL_HOMOGRAPHY = 4,
   GLOBAL_MOTION_TYPES
 } GLOBAL_MOTION_TYPE;
 
@@ -127,36 +143,53 @@ typedef struct {
 // regular motion vector) into a motion vector
 static INLINE int_mv gm_get_motion_vector(const Global_Motion_Params *gm) {
   int_mv res;
-  res.as_mv.row = (int16_t)ROUND_POWER_OF_TWO_SIGNED(gm->motion_params.wmmat[0],
+  res.as_mv.row = (int16_t)ROUND_POWER_OF_TWO_SIGNED(gm->motion_params.wmmat[1],
                                                      WARPEDMODEL_PREC_BITS - 3);
-  res.as_mv.col = (int16_t)ROUND_POWER_OF_TWO_SIGNED(gm->motion_params.wmmat[1],
+  res.as_mv.col = (int16_t)ROUND_POWER_OF_TWO_SIGNED(gm->motion_params.wmmat[0],
                                                      WARPEDMODEL_PREC_BITS - 3);
   return res;
 }
 
 static INLINE TransformationType gm_to_trans_type(GLOBAL_MOTION_TYPE gmtype) {
   switch (gmtype) {
-    case GLOBAL_ZERO: return UNKNOWN_TRANSFORM; break;
+    case GLOBAL_ZERO: return IDENTITY; break;
     case GLOBAL_TRANSLATION: return TRANSLATION; break;
     case GLOBAL_ROTZOOM: return ROTZOOM; break;
-    case GLOBAL_AFFINE: return AFFINE; break;
+    case GLOBAL_AFFINE:
+      return AFFINE;
+      break;
+    // case GLOBAL_HOMOGRAPHY: return HOMOGRAPHY; break;
     default: assert(0);
   }
   return UNKNOWN_TRANSFORM;
 }
 
 static INLINE GLOBAL_MOTION_TYPE get_gmtype(const Global_Motion_Params *gm) {
-  if (!gm->motion_params.wmmat[5] && !gm->motion_params.wmmat[4]) {
-    if (!gm->motion_params.wmmat[3] && !gm->motion_params.wmmat[2]) {
-      return ((!gm->motion_params.wmmat[1] && !gm->motion_params.wmmat[0])
-                  ? GLOBAL_ZERO
-                  : GLOBAL_TRANSLATION);
-    } else {
-      return GLOBAL_ROTZOOM;
-    }
-  } else {
-    return GLOBAL_AFFINE;
+  // if (gm->motion_params.wmmat[6] != 0 || gm->motion_params.wmmat[7] != 0)
+  //   return GLOBAL_HOMOGRAPHY;
+  if (gm->motion_params.wmmat[5] == (1 << WARPEDMODEL_PREC_BITS) &&
+      !gm->motion_params.wmmat[4] &&
+      gm->motion_params.wmmat[2] == (1 << WARPEDMODEL_PREC_BITS) &&
+      !gm->motion_params.wmmat[3]) {
+    return ((!gm->motion_params.wmmat[1] && !gm->motion_params.wmmat[0])
+                ? GLOBAL_ZERO
+                : GLOBAL_TRANSLATION);
   }
+  if (gm->motion_params.wmmat[2] == gm->motion_params.wmmat[5] &&
+      gm->motion_params.wmmat[3] == -gm->motion_params.wmmat[4])
+    return GLOBAL_ROTZOOM;
+  else
+    return GLOBAL_AFFINE;
+}
+
+static INLINE void set_default_gmparams(Global_Motion_Params *gm) {
+  static const int32_t default_gm_params[8] = {
+    0, 0, (1 << WARPEDMODEL_PREC_BITS), 0, 0, (1 << WARPEDMODEL_PREC_BITS), 0, 0
+  };
+  memcpy(gm->motion_params.wmmat, default_gm_params,
+         sizeof(gm->motion_params.wmmat));
+  gm->gmtype = GLOBAL_ZERO;
+  gm->motion_params.wmtype = IDENTITY;
 }
 #endif  // CONFIG_GLOBAL_MOTION
 

av1/common/reconinter.c

@@ -558,7 +558,7 @@ void build_inter_predictors(MACROBLOCKD *xd, int plane,
   for (ref = 0; ref < 1 + is_compound; ++ref) {
     gm[ref] = &xd->global_motion[mi->mbmi.ref_frame[ref]];
     is_global[ref] =
-        (get_y_mode(mi, block) == ZEROMV && get_gmtype(gm[ref]) > GLOBAL_ZERO);
+        (get_y_mode(mi, block) == ZEROMV && gm[ref]->gmtype > GLOBAL_ZERO);
   }
   // TODO(sarahparker) remove these once gm works with all experiments
   (void)gm;

av1/common/warped_motion.c

@@ -35,18 +35,18 @@ void project_points_translation(int32_t *mat, int *points, int *proj,
     const int x = *(points++), y = *(points++);
     if (subsampling_x)
       *(proj++) = ROUND_POWER_OF_TWO_SIGNED(
-          ((x * (1 << (WARPEDMODEL_PREC_BITS + 1))) + mat[1]),
+          ((x * (1 << (WARPEDMODEL_PREC_BITS + 1))) + mat[0]),
           WARPEDDIFF_PREC_BITS + 1);
     else
       *(proj++) = ROUND_POWER_OF_TWO_SIGNED(
-          ((x * (1 << WARPEDMODEL_PREC_BITS)) + mat[1]), WARPEDDIFF_PREC_BITS);
+          ((x * (1 << WARPEDMODEL_PREC_BITS)) + mat[0]), WARPEDDIFF_PREC_BITS);
     if (subsampling_y)
       *(proj++) = ROUND_POWER_OF_TWO_SIGNED(
-          ((y * (1 << (WARPEDMODEL_PREC_BITS + 1))) + mat[0]),
+          ((y * (1 << (WARPEDMODEL_PREC_BITS + 1))) + mat[1]),
           WARPEDDIFF_PREC_BITS + 1);
     else
       *(proj++) = ROUND_POWER_OF_TWO_SIGNED(
-          ((y * (1 << WARPEDMODEL_PREC_BITS))) + mat[0], WARPEDDIFF_PREC_BITS);
+          ((y * (1 << WARPEDMODEL_PREC_BITS))) + mat[1], WARPEDDIFF_PREC_BITS);
     points += stride_points - 2;
     proj += stride_proj - 2;
   }
@@ -60,19 +60,19 @@ void project_points_rotzoom(int32_t *mat, int *points, int *proj, const int n,
     const int x = *(points++), y = *(points++);
     if (subsampling_x)
       *(proj++) = ROUND_POWER_OF_TWO_SIGNED(
-          mat[3] * 2 * x + mat[2] * 2 * y + mat[1] +
-              (mat[3] + mat[2] - (1 << WARPEDMODEL_PREC_BITS)) / 2,
+          mat[2] * 2 * x + mat[3] * 2 * y + mat[0] +
+              (mat[2] + mat[3] - (1 << WARPEDMODEL_PREC_BITS)) / 2,
           WARPEDDIFF_PREC_BITS + 1);
     else
-      *(proj++) = ROUND_POWER_OF_TWO_SIGNED(mat[3] * x + mat[2] * y + mat[1],
+      *(proj++) = ROUND_POWER_OF_TWO_SIGNED(mat[2] * x + mat[3] * y + mat[0],
                                             WARPEDDIFF_PREC_BITS);
     if (subsampling_y)
       *(proj++) = ROUND_POWER_OF_TWO_SIGNED(
-          -mat[2] * 2 * x + mat[3] * 2 * y + mat[0] +
-              (-mat[2] + mat[3] - (1 << WARPEDMODEL_PREC_BITS)) / 2,
+          -mat[3] * 2 * x + mat[2] * 2 * y + mat[1] +
+              (-mat[3] + mat[2] - (1 << WARPEDMODEL_PREC_BITS)) / 2,
           WARPEDDIFF_PREC_BITS + 1);
     else
-      *(proj++) = ROUND_POWER_OF_TWO_SIGNED(-mat[2] * x + mat[3] * y + mat[0],
+      *(proj++) = ROUND_POWER_OF_TWO_SIGNED(-mat[3] * x + mat[2] * y + mat[1],
                                             WARPEDDIFF_PREC_BITS);
     points += stride_points - 2;
     proj += stride_proj - 2;
@@ -87,19 +87,19 @@ void project_points_affine(int32_t *mat, int *points, int *proj, const int n,
     const int x = *(points++), y = *(points++);
     if (subsampling_x)
       *(proj++) = ROUND_POWER_OF_TWO_SIGNED(
-          mat[3] * 2 * x + mat[2] * 2 * y + mat[1] +
-              (mat[3] + mat[2] - (1 << WARPEDMODEL_PREC_BITS)) / 2,
+          mat[2] * 2 * x + mat[3] * 2 * y + mat[0] +
+              (mat[2] + mat[3] - (1 << WARPEDMODEL_PREC_BITS)) / 2,
           WARPEDDIFF_PREC_BITS + 1);
     else
-      *(proj++) = ROUND_POWER_OF_TWO_SIGNED(mat[3] * x + mat[2] * y + mat[1],
+      *(proj++) = ROUND_POWER_OF_TWO_SIGNED(mat[2] * x + mat[3] * y + mat[0],
                                             WARPEDDIFF_PREC_BITS);
     if (subsampling_y)
       *(proj++) = ROUND_POWER_OF_TWO_SIGNED(
-          mat[4] * 2 * x + mat[5] * 2 * y + mat[0] +
+          mat[4] * 2 * x + mat[5] * 2 * y + mat[1] +
               (mat[4] + mat[5] - (1 << WARPEDMODEL_PREC_BITS)) / 2,
           WARPEDDIFF_PREC_BITS + 1);
     else
-      *(proj++) = ROUND_POWER_OF_TWO_SIGNED(mat[4] * x + mat[5] * y + mat[0],
+      *(proj++) = ROUND_POWER_OF_TWO_SIGNED(mat[4] * x + mat[5] * y + mat[1],
                                             WARPEDDIFF_PREC_BITS);
     points += stride_points - 2;
     proj += stride_proj - 2;
@@ -118,11 +118,11 @@ void project_points_homography(int32_t *mat, int *points, int *proj,
     x = (subsampling_x ? 4 * x + 1 : 2 * x);
     y = (subsampling_y ? 4 * y + 1 : 2 * y);
 
-    Z = (mat[7] * x + mat[6] * y + (1 << (WARPEDMODEL_ROW3HOMO_PREC_BITS + 1)));
-    xp = (mat[1] * x + mat[0] * y + 2 * mat[3]) *
+    Z = (mat[6] * x + mat[7] * y + (1 << (WARPEDMODEL_ROW3HOMO_PREC_BITS + 1)));
+    xp = (mat[2] * x + mat[3] * y + 2 * mat[0]) *
          (1 << (WARPEDPIXEL_PREC_BITS + WARPEDMODEL_ROW3HOMO_PREC_BITS -
                 WARPEDMODEL_PREC_BITS));
-    yp = (mat[2] * x + mat[5] * y + 2 * mat[4]) *
+    yp = (mat[4] * x + mat[5] * y + 2 * mat[1]) *
          (1 << (WARPEDPIXEL_PREC_BITS + WARPEDMODEL_ROW3HOMO_PREC_BITS -
                 WARPEDMODEL_PREC_BITS));
 
@@ -779,6 +779,7 @@ static int svdcmp(double **u, int m, int n, double w[], double **v) {
         break;
       }
       if (its == max_its - 1) {
+        aom_free(rv1);
         return 1;
       }
       assert(k > 0);
@@ -886,10 +887,16 @@ int pseudo_inverse(double *inv, double *matx, const int M, const int N) {
     return 1;
   }
   if (SVD(U, W, V, matx, M, N)) {
+    aom_free(U);
+    aom_free(W);
+    aom_free(V);
     return 1;
   }
   for (i = 0; i < N; i++) {
     if (fabs(W[i]) < TINY_NEAR_ZERO) {
+      aom_free(U);
+      aom_free(W);
+      aom_free(V);
       return 1;
     }
   }
@@ -978,12 +985,13 @@ static void denormalize_affine(double *params, double *T1, double *T2) {
   params_denorm[6] = params_denorm[7] = 0;
   params_denorm[8] = 1;
   denormalize_homography(params_denorm, T1, T2);
-  params[0] = params_denorm[5];
-  params[1] = params_denorm[2];
-  params[2] = params_denorm[1];
-  params[3] = params_denorm[0];
+  params[0] = params_denorm[2];
+  params[1] = params_denorm[5];
+  params[2] = params_denorm[0];
+  params[3] = params_denorm[1];
   params[4] = params_denorm[3];
   params[5] = params_denorm[4];
+  params[6] = params[7] = 0;
 }
 
 static void denormalize_rotzoom(double *params, double *T1, double *T2) {
@@ -997,10 +1005,13 @@ static void denormalize_rotzoom(double *params, double *T1, double *T2) {
   params_denorm[6] = params_denorm[7] = 0;
   params_denorm[8] = 1;
   denormalize_homography(params_denorm, T1, T2);
-  params[0] = params_denorm[5];
-  params[1] = params_denorm[2];
-  params[2] = params_denorm[1];
-  params[3] = params_denorm[0];
+  params[0] = params_denorm[2];
+  params[1] = params_denorm[5];
+  params[2] = params_denorm[0];
+  params[3] = params_denorm[1];
+  params[4] = -params[3];
+  params[5] = params[2];
+  params[6] = params[7] = 0;
 }
 
 static void denormalize_translation(double *params, double *T1, double *T2) {
@@ -1014,8 +1025,11 @@ static void denormalize_translation(double *params, double *T1, double *T2) {
   params_denorm[6] = params_denorm[7] = 0;
   params_denorm[8] = 1;
   denormalize_homography(params_denorm, T1, T2);
-  params[0] = params_denorm[5];
-  params[1] = params_denorm[2];
+  params[0] = params_denorm[2];
+  params[1] = params_denorm[5];
+  params[2] = params[5] = 1;
+  params[3] = params[4] = 0;
+  params[6] = params[7] = 0;
 }
 
 int find_translation(const int np, double *pts1, double *pts2, double *mat) {

av1/decoder/decodeframe.c

@@ -4010,27 +4010,32 @@ static void read_global_motion_params(Global_Motion_Params *params,
                                       aom_prob *probs, aom_reader *r) {
   GLOBAL_MOTION_TYPE gmtype =
       aom_read_tree(r, av1_global_motion_types_tree, probs, ACCT_STR);
+  set_default_gmparams(params);
   params->gmtype = gmtype;
   params->motion_params.wmtype = gm_to_trans_type(gmtype);
   switch (gmtype) {
     case GLOBAL_ZERO: break;
     case GLOBAL_AFFINE:
-      params->motion_params.wmmat[4] =
-          (aom_read_primitive_symmetric(r, GM_ABS_ALPHA_BITS) *
-           GM_ALPHA_DECODE_FACTOR);
-      params->motion_params.wmmat[5] =
-          aom_read_primitive_symmetric(r, GM_ABS_ALPHA_BITS) *
-              GM_ALPHA_DECODE_FACTOR +
-          (1 << WARPEDMODEL_PREC_BITS);
-    // fallthrough intended
     case GLOBAL_ROTZOOM:
       params->motion_params.wmmat[2] =
-          aom_read_primitive_symmetric(r, GM_ABS_ALPHA_BITS) *
-          GM_ALPHA_DECODE_FACTOR;
-      params->motion_params.wmmat[3] =
           (aom_read_primitive_symmetric(r, GM_ABS_ALPHA_BITS) *
            GM_ALPHA_DECODE_FACTOR) +
           (1 << WARPEDMODEL_PREC_BITS);
+      params->motion_params.wmmat[3] =
+          aom_read_primitive_symmetric(r, GM_ABS_ALPHA_BITS) *
+          GM_ALPHA_DECODE_FACTOR;
+      if (gmtype == GLOBAL_AFFINE) {
+        params->motion_params.wmmat[4] =
+            (aom_read_primitive_symmetric(r, GM_ABS_ALPHA_BITS) *
+             GM_ALPHA_DECODE_FACTOR);
+        params->motion_params.wmmat[5] =
+            aom_read_primitive_symmetric(r, GM_ABS_ALPHA_BITS) *
+                GM_ALPHA_DECODE_FACTOR +
+            (1 << WARPEDMODEL_PREC_BITS);
+      } else {
+        params->motion_params.wmmat[4] = -params->motion_params.wmmat[3];
+        params->motion_params.wmmat[5] = params->motion_params.wmmat[2];
+      }
     // fallthrough intended
     case GLOBAL_TRANSLATION:
       params->motion_params.wmmat[0] =
@@ -4046,18 +4051,17 @@ static void read_global_motion_params(Global_Motion_Params *params,
 
 static void read_global_motion(AV1_COMMON *cm, aom_reader *r) {
   int frame;
-  memset(cm->global_motion, 0, sizeof(cm->global_motion));
   for (frame = LAST_FRAME; frame <= ALTREF_FRAME; ++frame) {
     read_global_motion_params(&cm->global_motion[frame],
                               cm->fc->global_motion_types_prob, r);
     /*
-          printf("Dec Ref %d [%d]: %d %d %d %d\n",
-                 frame, cm->current_video_frame,
-                 cm->global_motion[frame].motion_params.wmmat[0].as_mv.row,
-                 cm->global_motion[frame].motion_params.wmmat[0].as_mv.col,
-                 cm->global_motion[frame].motion_params.wmmat[1].as_mv.row,
-                 cm->global_motion[frame].motion_params.wmmat[1].as_mv.col);
-    */
+    printf("Dec Ref %d [%d/%d]: %d %d %d %d\n",
+           frame, cm->current_video_frame, cm->show_frame,
+           cm->global_motion[frame].motion_params.wmmat[0],
+           cm->global_motion[frame].motion_params.wmmat[1],
+           cm->global_motion[frame].motion_params.wmmat[2],
+           cm->global_motion[frame].motion_params.wmmat[3]);
+           */
   }
 }
 #endif  // CONFIG_GLOBAL_MOTION

av1/encoder/bitstream.c

@@ -3952,28 +3952,29 @@ static void write_uncompressed_header(AV1_COMP *cpi,
 #if CONFIG_GLOBAL_MOTION
 static void write_global_motion_params(Global_Motion_Params *params,
                                        aom_prob *probs, aom_writer *w) {
-  GLOBAL_MOTION_TYPE gmtype = get_gmtype(params);
+  GLOBAL_MOTION_TYPE gmtype = params->gmtype;
   av1_write_token(w, av1_global_motion_types_tree, probs,
                   &global_motion_types_encodings[gmtype]);
   switch (gmtype) {
     case GLOBAL_ZERO: break;
     case GLOBAL_AFFINE:
-      aom_write_primitive_symmetric(
-          w, (params->motion_params.wmmat[4] >> GM_ALPHA_PREC_DIFF),
-          GM_ABS_ALPHA_BITS);
-      aom_write_primitive_symmetric(
-          w, (params->motion_params.wmmat[5] >> GM_ALPHA_PREC_DIFF) -
-                 (1 << GM_ALPHA_PREC_BITS),
-          GM_ABS_ALPHA_BITS);
-    // fallthrough intended
     case GLOBAL_ROTZOOM:
       aom_write_primitive_symmetric(
-          w, (params->motion_params.wmmat[2] >> GM_ALPHA_PREC_DIFF),
+          w, (params->motion_params.wmmat[2] >> GM_ALPHA_PREC_DIFF) -
+                 (1 << GM_ALPHA_PREC_BITS),
           GM_ABS_ALPHA_BITS);
       aom_write_primitive_symmetric(
-          w, (params->motion_params.wmmat[3] >> GM_ALPHA_PREC_DIFF) -
-                 (1 << GM_ALPHA_PREC_BITS),
+          w, (params->motion_params.wmmat[3] >> GM_ALPHA_PREC_DIFF),
           GM_ABS_ALPHA_BITS);
+      if (gmtype == GLOBAL_AFFINE) {
+        aom_write_primitive_symmetric(
+            w, (params->motion_params.wmmat[4] >> GM_ALPHA_PREC_DIFF),
+            GM_ABS_ALPHA_BITS);
+        aom_write_primitive_symmetric(
+            w, (params->motion_params.wmmat[5] >> GM_ALPHA_PREC_DIFF) -
+                   (1 << GM_ALPHA_PREC_BITS),
+            GM_ABS_ALPHA_BITS);
+      }
     // fallthrough intended
     case GLOBAL_TRANSLATION:
       aom_write_primitive_symmetric(
@@ -3992,18 +3993,19 @@ static void write_global_motion(AV1_COMP *cpi, aom_writer *w) {
   int frame;
   for (frame = LAST_FRAME; frame <= ALTREF_FRAME; ++frame) {
     if (!cpi->global_motion_used[frame]) {
-      memset(&cm->global_motion[frame], 0, sizeof(*cm->global_motion));
+      set_default_gmparams(&cm->global_motion[frame]);
     }
     write_global_motion_params(&cm->global_motion[frame],
                                cm->fc->global_motion_types_prob, w);
     /*
-          printf("Enc Ref %d [%d] (used %d): %d %d %d %d\n",
-                 frame, cm->current_video_frame, cpi->global_motion_used[frame],
-                 cm->global_motion[frame].motion_params.wmmat[0].as_mv.row,
-                 cm->global_motion[frame].motion_params.wmmat[0].as_mv.col,
-                 cm->global_motion[frame].motion_params.wmmat[1].as_mv.row,
-                 cm->global_motion[frame].motion_params.wmmat[1].as_mv.col);
-    */
+    printf("Enc Ref %d [%d/%d] (used %d): %d %d %d %d\n",
+           frame, cm->current_video_frame, cm->show_frame,
+           cpi->global_motion_used[frame],
+           cm->global_motion[frame].motion_params.wmmat[0],
+           cm->global_motion[frame].motion_params.wmmat[1],
+           cm->global_motion[frame].motion_params.wmmat[2],
+           cm->global_motion[frame].motion_params.wmmat[3]);
+           */
   }
 }
 #endif

av1/encoder/encodeframe.c

@@ -4612,7 +4612,7 @@ static int input_fpmb_stats(FIRSTPASS_MB_STATS *firstpass_mb_stats,
 #endif
 
 #if CONFIG_GLOBAL_MOTION
-#define MIN_TRANS_THRESH 8
+#define MIN_TRANS_THRESH (1 * GM_TRANS_DECODE_FACTOR)
 #define GLOBAL_MOTION_ADVANTAGE_THRESH 0.60
 #define GLOBAL_MOTION_MODEL ROTZOOM
 
@@ -4621,24 +4621,26 @@ static int input_fpmb_stats(FIRSTPASS_MB_STATS *firstpass_mb_stats,
 // zero-centering.
 static int32_t add_param_offset(int param_index, int32_t param_value,
                                 int32_t offset) {
-  const int scale_vals[2] = { GM_ALPHA_PREC_DIFF, GM_TRANS_PREC_DIFF };
-  const int clamp_vals[2] = { GM_ALPHA_MAX, GM_TRANS_MAX };
-  const int is_trans_param = param_index < 2;
-  const int is_one_centered = (!is_trans_param) && (param_index & 1);
+  const int scale_vals[3] = { GM_TRANS_PREC_DIFF, GM_ALPHA_PREC_DIFF,
+                              GM_ROW3HOMO_PREC_DIFF };
+  const int clamp_vals[3] = { GM_TRANS_MAX, GM_ALPHA_MAX, GM_ROW3HOMO_MAX };
+  // type of param: 0 - translation, 1 - affine, 2 - homography
+  const int param_type = (param_index < 2 ? 0 : (param_index < 6 ? 1 : 2));
+  const int is_one_centered = (param_index == 2 || param_index == 5);
 
   // Make parameter zero-centered and offset the shift that was done to make
   // it compatible with the warped model
   param_value = (param_value - (is_one_centered << WARPEDMODEL_PREC_BITS)) >>
-                scale_vals[is_trans_param];
+                scale_vals[param_type];
   // Add desired offset to the rescaled/zero-centered parameter
   param_value += offset;
   // Clamp the parameter so it does not overflow the number of bits allotted
   // to it in the bitstream
-  param_value = (int32_t)clamp(param_value, -clamp_vals[is_trans_param],
-                               clamp_vals[is_trans_param]);
+  param_value = (int32_t)clamp(param_value, -clamp_vals[param_type],
+                               clamp_vals[param_type]);
   // Rescale the parameter to WARPEDMODEL_PRECISION_BITS so it is compatible
   // with the warped motion library
-  param_value *= (1 << scale_vals[is_trans_param]);
+  param_value *= (1 << scale_vals[param_type]);
 
   // Undo the zero-centering step if necessary
   return param_value + (is_one_centered << WARPEDMODEL_PREC_BITS);
@@ -4712,16 +4714,18 @@ static void refine_integerized_param(WarpedMotionParams *wm,
       // step is too wide
       step >>= 1;
     }
-
     *param = best_param;
   }
+  // For rotzoom model enforce the constraints on mat[4] and mat[5]
+  if (wm->wmtype == ROTZOOM) {
+    param_mat[5] = param_mat[2];
+    param_mat[4] = -param_mat[3];
+  }
 }
 
-static void convert_to_params(const double *params, TransformationType type,
-                              int32_t *model) {
-  int i, diag_value;
+static void convert_to_params(const double *params, int32_t *model) {
+  int i;
   int alpha_present = 0;
-  int n_params = n_trans_model_params[type];
   model[0] = (int32_t)floor(params[0] * (1 << GM_TRANS_PREC_BITS) + 0.5);
   model[1] = (int32_t)floor(params[1] * (1 << GM_TRANS_PREC_BITS) + 0.5);
   model[0] = (int32_t)clamp(model[0], GM_TRANS_MIN, GM_TRANS_MAX) *
@@ -4729,13 +4733,18 @@ static void convert_to_params(const double *params, TransformationType type,
   model[1] = (int32_t)clamp(model[1], GM_TRANS_MIN, GM_TRANS_MAX) *
              GM_TRANS_DECODE_FACTOR;
 
-  for (i = 2; i < n_params; ++i) {
-    diag_value = ((i & 1) ? (1 << GM_ALPHA_PREC_BITS) : 0);
+  for (i = 2; i < 6; ++i) {
+    const int diag_value = ((i == 2 || i == 5) ? (1 << GM_ALPHA_PREC_BITS) : 0);
     model[i] = (int32_t)floor(params[i] * (1 << GM_ALPHA_PREC_BITS) + 0.5);
     model[i] =
-        (int32_t)(clamp(model[i] - diag_value, GM_ALPHA_MIN, GM_ALPHA_MAX) +
-                  diag_value) *
-        GM_ALPHA_DECODE_FACTOR;
+        (int32_t)clamp(model[i] - diag_value, GM_ALPHA_MIN, GM_ALPHA_MAX);
+    alpha_present |= (model[i] != 0);
+    model[i] = (model[i] + diag_value) * GM_ALPHA_DECODE_FACTOR;
+  }
+  for (; i < 8; ++i) {
+    model[i] = (int32_t)floor(params[i] * (1 << GM_ROW3HOMO_PREC_BITS) + 0.5);
+    model[i] = (int32_t)clamp(model[i], GM_ROW3HOMO_MIN, GM_ROW3HOMO_MAX) *
+               GM_ROW3HOMO_DECODE_FACTOR;
     alpha_present |= (model[i] != 0);
   }
 
@@ -4748,11 +4757,8 @@ static void convert_to_params(const double *params, TransformationType type,
 }
 
 static void convert_model_to_params(const double *params,
-                                    TransformationType type,
                                     Global_Motion_Params *model) {
-  // TODO(sarahparker) implement for homography
-  if (type > HOMOGRAPHY)
-    convert_to_params(params, type, model->motion_params.wmmat);
+  convert_to_params(params, model->motion_params.wmmat);
   model->gmtype = get_gmtype(model);
   model->motion_params.wmtype = gm_to_trans_type(model->gmtype);
 }
@@ -4782,19 +4788,21 @@ static void encode_frame_internal(AV1_COMP *cpi) {
 #if CONFIG_GLOBAL_MOTION
   aom_clear_system_state();
   av1_zero(cpi->global_motion_used);
+  for (i = LAST_FRAME; i <= ALTREF_FRAME; ++i) {
+    set_default_gmparams(&cm->global_motion[i]);
+  }
   if (cpi->common.frame_type == INTER_FRAME && cpi->Source) {
     YV12_BUFFER_CONFIG *ref_buf;
     int frame;
     double erroradvantage = 0;
-    double params[9] = { 0, 0, 0, 0, 0, 0, 0, 0, 0 };
+    double params[8] = { 0, 0, 1, 0, 0, 1, 0, 0 };
     for (frame = LAST_FRAME; frame <= ALTREF_FRAME; ++frame) {
       ref_buf = get_ref_frame_buffer(cpi, frame);
       if (ref_buf) {
         if (compute_global_motion_feature_based(GLOBAL_MOTION_MODEL,
                                                 cpi->Source, ref_buf, params)) {
-          convert_model_to_params(params, GLOBAL_MOTION_MODEL,
-                                  &cm->global_motion[frame]);
-          if (get_gmtype(&cm->global_motion[frame]) > GLOBAL_ZERO) {
+          convert_model_to_params(params, &cm->global_motion[frame]);
+          if (cm->global_motion[frame].gmtype > GLOBAL_ZERO) {
             refine_integerized_param(
                 &cm->global_motion[frame].motion_params,
 #if CONFIG_AOM_HIGHBITDEPTH
@@ -4814,9 +4822,8 @@ static void encode_frame_internal(AV1_COMP *cpi) {
                 cpi->Source->y_width, cpi->Source->y_height,
                 cpi->Source->y_stride, 0, 0, 16, 16);
             if (erroradvantage > GLOBAL_MOTION_ADVANTAGE_THRESH)
-              // Not enough advantage in using a global model. Make 0.
-              memset(&cm->global_motion[frame], 0,
-                     sizeof(cm->global_motion[frame]));
+              // Not enough advantage in using a global model. Make identity.
+              set_default_gmparams(&cm->global_motion[frame]);
           }
         }
       }

av1/encoder/ransac.c

@@ -41,8 +41,8 @@ static void project_points_double_translation(double *mat, double *points,
   int i;
   for (i = 0; i < n; ++i) {
     const double x = *(points++), y = *(points++);
-    *(proj++) = x + mat[1];
-    *(proj++) = y + mat[0];
+    *(proj++) = x + mat[0];
+    *(proj++) = y + mat[1];
     points += stride_points - 2;
     proj += stride_proj - 2;
   }
@@ -55,8 +55,8 @@ static void project_points_double_rotzoom(double *mat, double *points,