EEEVEE: Object Motion Blur: Initial Implementation

This adds object motion blur vectors for EEVEE as well as better noise
reduction for it.

For TAA reprojection we just compute the motion vector on the fly based on
camera motion and depth buffer. This makes possible to store another motion
vector only for the blurring which is not useful for TAA history fetching.

Motion Data is saved per object & per geometry if using deformation blur.
We support deformation motion blur by saving previous VBO and modifying the
actual GPUBatch for the geometry to include theses VBOs.

We store Previous and Next frame motion in the same motion vector buffer
(RG for prev and BA for next). This makes non linear motion blur (like
rotating objects) less prone to outward/inward blur.

We also improve the motion blur post process to expand outside the objects
border. We use a tile base approach and the max size of the blur is set via
a new render setting.

We use a background reconstruction method that needs another setting
(Background Separation).

Sampling is done using a fixed 8 dithered samples per direction. The final
render samples will clear the noise like other stochastic effects.

One caveat is that hair particles are not yet supported. Support will
come in another patch.

Reviewed By: jbakker

Differential Revision: https://developer.blender.org/D7297

1 files changed, 209 insertions, 38 deletions

diff --git a/source/blender/draw/engines/eevee/shaders/effect_motion_blur_frag.glsl b/source/blender/draw/engines/eevee/shaders/effect_motion_blur_frag.glsl
index b7935235d06..fbf507a2e40 100644
--- a/source/blender/draw/engines/eevee/shaders/effect_motion_blur_frag.glsl
+++ b/source/blender/draw/engines/eevee/shaders/effect_motion_blur_frag.glsl
@@ -1,64 +1,235 @@
 
+/*
+ * Based on:
+ * A Fast and Stable Feature-Aware Motion Blur Filter
+ * by Jean-Philippe Guertin, Morgan McGuire, Derek Nowrouzezahrai
+ *
+ * With modification from the presentation:
+ * Next Generation Post Processing in Call of Duty Advanced Warfare
+ * by Jorge Jimenez
+ */
 uniform sampler2D colorBuffer;
 uniform sampler2D depthBuffer;
+uniform sampler2D velocityBuffer;
+uniform sampler2D tileMaxBuffer;
 
-/* current frame */
-uniform mat4 currInvViewProjMatrix;
+#define KERNEL 8
 
-/* past frame frame */
-uniform mat4 pastViewProjMatrix;
+/* TODO(fclem) deduplicate this code. */
+uniform sampler2DArray utilTex;
+#define LUT_SIZE 64
+#define texelfetch_noise_tex(coord) texelFetch(utilTex, ivec3(ivec2(coord) % LUT_SIZE, 2.0), 0)
+
+uniform float depthScale;
+uniform ivec2 tileBufferSize;
+uniform vec2 viewportSize;
+uniform vec2 viewportSizeInv;
+uniform bool isPerspective;
+uniform vec2 nearFar; /* Near & far view depths values */
+
+#define linear_depth(z) \
+  ((isPerspective) ? (nearFar.x * nearFar.y) / (z * (nearFar.x - nearFar.y) + nearFar.y) : \
+                     z * (nearFar.y - nearFar.x) + nearFar.x) /* Only true for camera view! */
 
 in vec4 uvcoordsvar;
 
-out vec4 FragColor;
+out vec4 fragColor;
 
-#define MAX_SAMPLE 64
+#define saturate(a) clamp(a, 0.0, 1.0)
 
-uniform int samples;
+vec2 spread_compare(float center_motion_length, float sample_motion_length, float offset_length)
+{
+  return saturate(vec2(center_motion_length, sample_motion_length) - offset_length + 1.0);
+}
 
-float wang_hash_noise(uint s)
+vec2 depth_compare(float center_depth, float sample_depth)
 {
-  uint seed = (uint(gl_FragCoord.x) * 1664525u + uint(gl_FragCoord.y)) + s;
+  return saturate(0.5 + vec2(depthScale, -depthScale) * (sample_depth - center_depth));
+}
 
-  seed = (seed ^ 61u) ^ (seed >> 16u);
-  seed *= 9u;
-  seed = seed ^ (seed >> 4u);
-  seed *= 0x27d4eb2du;
-  seed = seed ^ (seed >> 15u);
+/* Kill contribution if not going the same direction. */
+float dir_compare(vec2 offset, vec2 sample_motion, float sample_motion_length)
+{
+  if (sample_motion_length < 0.5) {
+    return 1.0;
+  }
+  return (dot(offset, sample_motion) > 0.0) ? 1.0 : 0.0;
+}
 
-  float value = float(seed);
-  value *= 1.0 / 4294967296.0;
-  return fract(value);
+/* Return background (x) and foreground (y) weights. */
+vec2 sample_weights(float center_depth,
+                    float sample_depth,
+                    float center_motion_length,
+                    float sample_motion_length,
+                    float offset_length)
+{
+  /* Clasify foreground/background. */
+  vec2 depth_weight = depth_compare(center_depth, sample_depth);
+  /* Weight if sample is overlapping or under the center pixel. */
+  vec2 spread_weight = spread_compare(center_motion_length, sample_motion_length, offset_length);
+  return depth_weight * spread_weight;
 }
 
-void main()
+vec4 decode_velocity(vec4 velocity)
 {
-  vec3 ndc_pos;
-  ndc_pos.xy = uvcoordsvar.xy;
-  ndc_pos.z = texture(depthBuffer, uvcoordsvar.xy).x;
+  velocity = velocity * 2.0 - 1.0;
+  /* Needed to match cycles. Can't find why... (fclem) */
+  velocity *= 0.5;
+  /* Transpose to pixelspace. */
+  velocity *= viewportSize.xyxy;
+  return velocity;
+}
 
-  float inv_samples = 1.0 / float(samples);
-  float noise = 2.0 * wang_hash_noise(0u) * inv_samples;
+vec4 sample_velocity(vec2 uv)
+{
+  vec4 data = texture(velocityBuffer, uv);
+  return decode_velocity(data);
+}
 
-  /* Normalize Device Coordinates are [-1, +1]. */
-  ndc_pos = ndc_pos * 2.0 - 1.0;
+vec2 sample_velocity(vec2 uv, const bool next)
+{
+  vec4 data = sample_velocity(uv);
+  data.xy = (next ? data.zw : data.xy);
+  return data.xy;
+}
 
-  vec4 p = currInvViewProjMatrix * vec4(ndc_pos, 1.0);
-  vec3 world_pos = p.xyz / p.w; /* Perspective divide */
+void gather_sample(vec2 screen_uv,
+                   float center_depth,
+                   float center_motion_len,
+                   vec2 offset,
+                   float offset_len,
+                   const bool next,
+                   inout vec4 accum,
+                   inout vec4 accum_bg,
+                   inout vec3 w_accum)
+{
+  vec2 sample_uv = screen_uv - offset * viewportSizeInv;
+  vec2 sample_motion = sample_velocity(sample_uv, next);
+  float sample_motion_len = length(sample_motion);
+  float sample_depth = linear_depth(texture(depthBuffer, sample_uv).r);
+  vec4 col = textureLod(colorBuffer, sample_uv, 0.0);
+
+  vec3 weights;
+  weights.xy = sample_weights(
+      center_depth, sample_depth, center_motion_len, sample_motion_len, offset_len);
+  weights.z = dir_compare(offset, sample_motion, sample_motion_len);
+  weights.xy *= weights.z;
+
+  accum += col * weights.y;
+  accum_bg += col * weights.x;
+  w_accum += weights;
+}
 
-  /* Now find where was this pixel position
-   * inside the past camera viewport */
-  vec4 old_ndc = pastViewProjMatrix * vec4(world_pos, 1.0);
-  old_ndc.xyz /= old_ndc.w; /* Perspective divide */
+void gather_blur(vec2 screen_uv,
+                 vec2 center_motion,
+                 float center_depth,
+                 vec2 max_motion,
+                 float ofs,
+                 const bool next,
+                 inout vec4 accum,
+                 inout vec4 accum_bg,
+                 inout vec3 w_accum)
+{
+  float center_motion_len = length(center_motion);
+  float max_motion_len = length(max_motion);
+
+  /* Tile boundaries randomization can fetch a tile where there is less motion than this pixel.
+   * Fix this by overriding the max_motion. */
+  if (max_motion_len < center_motion_len) {
+    max_motion_len = center_motion_len;
+    max_motion = center_motion;
+  }
 
-  vec2 motion = (ndc_pos.xy - old_ndc.xy) * 0.25; /* 0.25 fit cycles ref */
+  if (max_motion_len < 0.5) {
+    return;
+  }
 
-  float inc = 2.0 * inv_samples;
-  float i = -1.0 + noise;
+  int i;
+  float t, inc = 1.0 / float(KERNEL);
+  for (i = 0, t = ofs * inc; i < KERNEL; i++, t += inc) {
+    gather_sample(screen_uv,
+                  center_depth,
+                  center_motion_len,
+                  max_motion * t,
+                  max_motion_len * t,
+                  next,
+                  accum,
+                  accum_bg,
+                  w_accum);
+  }
 
-  FragColor = vec4(0.0);
-  for (int j = 0; j < samples && j < MAX_SAMPLE; j++) {
-    FragColor += textureLod(colorBuffer, uvcoordsvar.xy + motion * i, 0.0) * inv_samples;
-    i += inc;
+  if (center_motion_len < 0.5) {
+    return;
   }
+
+  for (i = 0, t = ofs * inc; i < KERNEL; i++, t += inc) {
+    /* Also sample in center motion direction.
+     * Allow to recover motion where there is conflicting
+     * motion between foreground and background. */
+    gather_sample(screen_uv,
+                  center_depth,
+                  center_motion_len,
+                  center_motion * t,
+                  center_motion_len * t,
+                  next,
+                  accum,
+                  accum_bg,
+                  w_accum);
+  }
+}
+
+void main()
+{
+  vec2 uv = uvcoordsvar.xy;
+
+  /* Data of the center pixel of the gather (target). */
+  float center_depth = linear_depth(texture(depthBuffer, uv).r);
+  vec4 center_motion = sample_velocity(uv);
+  vec4 center_color = textureLod(colorBuffer, uv, 0.0);
+
+  vec2 rand = texelfetch_noise_tex(gl_FragCoord.xy).xy;
+
+  /* Randomize tile boundary to avoid ugly discontinuities. Randomize 1/4th of the tile.
+   * Note this randomize only in one direction but in practice it's enough. */
+  rand.x = rand.x * 2.0 - 1.0;
+  ivec2 tile = ivec2(gl_FragCoord.xy + rand.x * float(EEVEE_VELOCITY_TILE_SIZE) * 0.25) /
+               EEVEE_VELOCITY_TILE_SIZE;
+  tile = clamp(tile, ivec2(0), tileBufferSize - 1);
+  vec4 max_motion = decode_velocity(texelFetch(tileMaxBuffer, tile, 0));
+
+  /* First (center) sample: time = T */
+  /* x: Background, y: Foreground, z: dir. */
+  vec3 w_accum = vec3(0.0, 0.0, 1.0);
+  vec4 accum_bg = vec4(0.0);
+  vec4 accum = vec4(0.0);
+  /* First linear gather. time = [T - delta, T] */
+  gather_blur(
+      uv, center_motion.xy, center_depth, max_motion.xy, rand.y, false, accum, accum_bg, w_accum);
+  /* Second linear gather. time = [T, T + delta] */
+  gather_blur(
+      uv, center_motion.zw, center_depth, max_motion.zw, rand.y, true, accum, accum_bg, w_accum);
+
+#if 1
+  /* Avoid division by 0.0. */
+  float w = 1.0 / (50.0 * float(KERNEL) * 4.0);
+  accum_bg += center_color * w;
+  w_accum.x += w;
+  /* Note: In Jimenez's presentation, they used center sample.
+   * We use background color as it contains more informations for foreground
+   * elements that have not enough weights.
+   * Yield beter blur in complex motion. */
+  center_color = accum_bg / w_accum.x;
+#endif
+  /* Merge background. */
+  accum += accum_bg;
+  w_accum.y += w_accum.x;
+  /* Balance accumulation for failled samples.
+   * We replace the missing foreground by the background. */
+  float blend_fac = saturate(1.0 - w_accum.y / w_accum.z);
+  fragColor = (accum / w_accum.z) + center_color * blend_fac;
+
+#if 0 /* For debugging. */
+  fragColor.rgb = fragColor.ggg;
+  fragColor.rg += max_motion.xy;
+#endif
 }