Cycles Denoising: Merge outlier heuristic and confidence interval test

The previous outlier heuristic only checked whether the pixel is more than
twice as bright compared to the 75% quantile of the 5x5 neighborhood.
While this detected fireflies robustly, it also incorrectly marked a lot of
legitimate small highlights as outliers and filtered them away.

This commit adds an additional condition for marking a pixel as a firefly:
In addition to being above the reference brightness, the lower end of the
3-sigma confidence interval has to be below it.
Since the lower end approximates how low the true value of the pixel might be,
this test separates pixels that are supposed to be very bright from pixels that
are very bright due to random fireflies.

Also, since there is now a reliable outlier filter as a preprocessing step,
the additional confidence interval test in the reconstruction kernel is no
longer needed.

6 files changed, 13 insertions, 31 deletions

diff --git a/intern/cycles/device/device_cpu.cpp b/intern/cycles/device/device_cpu.cpp
index c2f74aa8903..29bb1f91a40 100644
--- a/intern/cycles/device/device_cpu.cpp
+++ b/intern/cycles/device/device_cpu.cpp
@@ -185,9 +185,9 @@ public:
 	KernelFunctions<void(*)(int, int, float*, float*, float*, float*, int*, int, int)>       filter_nlm_update_output_kernel;
 	KernelFunctions<void(*)(float*, float*, int*, int)>                                      filter_nlm_normalize_kernel;
 
-	KernelFunctions<void(*)(float*, int, int, int, float*, int*, int*, int, int, float)>                                              filter_construct_transform_kernel;
-	KernelFunctions<void(*)(int, int, float*, float*, float*, float*, float*, int*, float*, float3*, int*, int*, int, int, int, int)> filter_nlm_construct_gramian_kernel;
-	KernelFunctions<void(*)(int, int, int, int, int, float*, int*, float*, float3*, int*, int)>                                       filter_finalize_kernel;
+	KernelFunctions<void(*)(float*, int, int, int, float*, int*, int*, int, int, float)>                              filter_construct_transform_kernel;
+	KernelFunctions<void(*)(int, int, float*, float*, float*, int*, float*, float3*, int*, int*, int, int, int, int)> filter_nlm_construct_gramian_kernel;
+	KernelFunctions<void(*)(int, int, int, int, int, float*, int*, float*, float3*, int*, int)>                       filter_finalize_kernel;
 
 	KernelFunctions<void(*)(KernelGlobals *, ccl_constant KernelData*, ccl_global void*, int, ccl_global char*,
 	                       ccl_global uint*, int, int, int, int, int, int, int, int, ccl_global int*, int,
@@ -465,8 +465,6 @@ public:
 
 	bool denoising_reconstruct(device_ptr color_ptr,
 	                           device_ptr color_variance_ptr,
-	                           device_ptr guide_ptr,
-	                           device_ptr guide_variance_ptr,
 	                           device_ptr output_ptr,
 	                           DenoisingTask *task)
 	{
@@ -485,8 +483,8 @@ public:
 			                     task->reconstruction_state.source_w - max(0, dx),
 			                     task->reconstruction_state.source_h - max(0, dy)};
 			filter_nlm_calc_difference_kernel()(dx, dy,
-			                                    (float*) guide_ptr,
-			                                    (float*) guide_variance_ptr,
+			                                    (float*) color_ptr,
+			                                    (float*) color_variance_ptr,
 			                                    difference,
 			                                    local_rect,
 			                                    task->buffer.w,
@@ -499,8 +497,6 @@ public:
 			filter_nlm_construct_gramian_kernel()(dx, dy,
 			                                      blurDifference,
 			                                      (float*)  task->buffer.mem.device_pointer,
-			                                      (float*)  color_ptr,
-			                                      (float*)  color_variance_ptr,
 			                                      (float*)  task->storage.transform.device_pointer,
 			                                      (int*)    task->storage.rank.device_pointer,
 			                                      (float*)  task->storage.XtWX.device_pointer,
@@ -648,7 +644,7 @@ public:
 		DenoisingTask denoising(this);
 
 		denoising.functions.construct_transform = function_bind(&CPUDevice::denoising_construct_transform, this, &denoising);
-		denoising.functions.reconstruct = function_bind(&CPUDevice::denoising_reconstruct, this, _1, _2, _3, _4, _5, &denoising);
+		denoising.functions.reconstruct = function_bind(&CPUDevice::denoising_reconstruct, this, _1, _2, _3, &denoising);
 		denoising.functions.divide_shadow = function_bind(&CPUDevice::denoising_divide_shadow, this, _1, _2, _3, _4, _5, &denoising);
 		denoising.functions.non_local_means = function_bind(&CPUDevice::denoising_non_local_means, this, _1, _2, _3, _4, &denoising);
 		denoising.functions.combine_halves = function_bind(&CPUDevice::denoising_combine_halves, this, _1, _2, _3, _4, _5, _6, &denoising);
diff --git a/intern/cycles/device/device_cuda.cpp b/intern/cycles/device/device_cuda.cpp
index 99537e9a983..29fa08d94b1 100644
--- a/intern/cycles/device/device_cuda.cpp
+++ b/intern/cycles/device/device_cuda.cpp
@@ -1051,8 +1051,6 @@ public:
 
 	bool denoising_reconstruct(device_ptr color_ptr,
 	                           device_ptr color_variance_ptr,
-	                           device_ptr guide_ptr,
-	                           device_ptr guide_variance_ptr,
 	                           device_ptr output_ptr,
 	                           DenoisingTask *task)
 	{
@@ -1096,8 +1094,8 @@ public:
 			                     task->reconstruction_state.source_h - max(0, dy)};
 
 			void *calc_difference_args[] = {&dx, &dy,
-			                                &guide_ptr,
-			                                &guide_variance_ptr,
+			                                &color_ptr,
+			                                &color_variance_ptr,
 			                                &difference,
 			                                &local_rect,
 			                                &task->buffer.w,
@@ -1126,8 +1124,6 @@ public:
 			void *construct_gramian_args[] = {&dx, &dy,
 			                                  &blurDifference,
 			                                  &task->buffer.mem.device_pointer,
-			                                  &color_ptr,
-			                                  &color_variance_ptr,
 			                                  &task->storage.transform.device_pointer,
 			                                  &task->storage.rank.device_pointer,
 			                                  &task->storage.XtWX.device_pointer,
@@ -1294,7 +1290,7 @@ public:
 		DenoisingTask denoising(this);
 
 		denoising.functions.construct_transform = function_bind(&CUDADevice::denoising_construct_transform, this, &denoising);
-		denoising.functions.reconstruct = function_bind(&CUDADevice::denoising_reconstruct, this, _1, _2, _3, _4, _5, &denoising);
+		denoising.functions.reconstruct = function_bind(&CUDADevice::denoising_reconstruct, this, _1, _2, _3, &denoising);
 		denoising.functions.divide_shadow = function_bind(&CUDADevice::denoising_divide_shadow, this, _1, _2, _3, _4, _5, &denoising);
 		denoising.functions.non_local_means = function_bind(&CUDADevice::denoising_non_local_means, this, _1, _2, _3, _4, &denoising);
 		denoising.functions.combine_halves = function_bind(&CUDADevice::denoising_combine_halves, this, _1, _2, _3, _4, _5, _6, &denoising);
diff --git a/intern/cycles/device/device_denoising.cpp b/intern/cycles/device/device_denoising.cpp
index 613bd9112cf..619cc1d171e 100644
--- a/intern/cycles/device/device_denoising.cpp
+++ b/intern/cycles/device/device_denoising.cpp
@@ -215,7 +215,7 @@ bool DenoisingTask::run_denoising()
 	{
 		device_sub_ptr color_ptr    (device, buffer.mem,  8*buffer.pass_stride, 3*buffer.pass_stride, MEM_READ_WRITE);
 		device_sub_ptr color_var_ptr(device, buffer.mem, 11*buffer.pass_stride, 3*buffer.pass_stride, MEM_READ_WRITE);
-		functions.reconstruct(*color_ptr, *color_var_ptr, *color_ptr, *color_var_ptr, render_buffer.ptr);
+		functions.reconstruct(*color_ptr, *color_var_ptr, render_buffer.ptr);
 	}
 
 	device->mem_free(storage.XtWX);
diff --git a/intern/cycles/device/device_denoising.h b/intern/cycles/device/device_denoising.h
index 25b93c2ad74..def7b72f67d 100644
--- a/intern/cycles/device/device_denoising.h
+++ b/intern/cycles/device/device_denoising.h
@@ -58,8 +58,6 @@ public:
 		              )> non_local_means;
 		function<bool(device_ptr color_ptr,
 		              device_ptr color_variance_ptr,
-		              device_ptr guide_ptr,
-		              device_ptr guide_variance_ptr,
 		              device_ptr output_ptr
 		              )> reconstruct;
 		function<bool()> construct_transform;
diff --git a/intern/cycles/device/opencl/opencl.h b/intern/cycles/device/opencl/opencl.h
index 27e196d1e68..52851061d7b 100644
--- a/intern/cycles/device/opencl/opencl.h
+++ b/intern/cycles/device/opencl/opencl.h
@@ -390,8 +390,6 @@ protected:
 	bool denoising_construct_transform(DenoisingTask *task);
 	bool denoising_reconstruct(device_ptr color_ptr,
 	                           device_ptr color_variance_ptr,
-	                           device_ptr guide_ptr,
-	                           device_ptr guide_variance_ptr,
 	                           device_ptr output_ptr,
 	                           DenoisingTask *task);
 	bool denoising_combine_halves(device_ptr a_ptr,
diff --git a/intern/cycles/device/opencl/opencl_base.cpp b/intern/cycles/device/opencl/opencl_base.cpp
index 24b70e3446c..e4ab979dcbf 100644
--- a/intern/cycles/device/opencl/opencl_base.cpp
+++ b/intern/cycles/device/opencl/opencl_base.cpp
@@ -693,8 +693,6 @@ bool OpenCLDeviceBase::denoising_construct_transform(DenoisingTask *task)
 
 bool OpenCLDeviceBase::denoising_reconstruct(device_ptr color_ptr,
                                              device_ptr color_variance_ptr,
-                                             device_ptr guide_ptr,
-                                             device_ptr guide_variance_ptr,
                                              device_ptr output_ptr,
                                              DenoisingTask *task)
 {
@@ -703,8 +701,6 @@ bool OpenCLDeviceBase::denoising_reconstruct(device_ptr color_ptr,
 
 	cl_mem color_mem = CL_MEM_PTR(color_ptr);
 	cl_mem color_variance_mem = CL_MEM_PTR(color_variance_ptr);
-	cl_mem guide_mem = CL_MEM_PTR(guide_ptr);
-	cl_mem guide_variance_mem = CL_MEM_PTR(guide_variance_ptr);
 	cl_mem output_mem = CL_MEM_PTR(output_ptr);
 
 	cl_mem buffer_mem = CL_MEM_PTR(task->buffer.mem.device_pointer);
@@ -735,8 +731,8 @@ bool OpenCLDeviceBase::denoising_reconstruct(device_ptr color_ptr,
 
 		kernel_set_args(ckNLMCalcDifference, 0,
 		                dx, dy,
-		                guide_mem,
-		                guide_variance_mem,
+		                color_mem,
+		                color_variance_mem,
 		                difference,
 		                local_rect,
 		                task->buffer.w,
@@ -775,8 +771,6 @@ bool OpenCLDeviceBase::denoising_reconstruct(device_ptr color_ptr,
 		                dx, dy,
 		                blurDifference,
 		                buffer_mem,
-		                color_mem,
-		                color_variance_mem,
 		                transform_mem,
 		                rank_mem,
 		                XtWX_mem,
@@ -961,7 +955,7 @@ void OpenCLDeviceBase::denoise(RenderTile &rtile, const DeviceTask &task)
 
 	denoising.functions.set_tiles = function_bind(&OpenCLDeviceBase::denoising_set_tiles, this, _1, &denoising);
 	denoising.functions.construct_transform = function_bind(&OpenCLDeviceBase::denoising_construct_transform, this, &denoising);
-	denoising.functions.reconstruct = function_bind(&OpenCLDeviceBase::denoising_reconstruct, this, _1, _2, _3, _4, _5, &denoising);
+	denoising.functions.reconstruct = function_bind(&OpenCLDeviceBase::denoising_reconstruct, this, _1, _2, _3, &denoising);
 	denoising.functions.divide_shadow = function_bind(&OpenCLDeviceBase::denoising_divide_shadow, this, _1, _2, _3, _4, _5, &denoising);
 	denoising.functions.non_local_means = function_bind(&OpenCLDeviceBase::denoising_non_local_means, this, _1, _2, _3, _4, &denoising);
 	denoising.functions.combine_halves = function_bind(&OpenCLDeviceBase::denoising_combine_halves, this, _1, _2, _3, _4, _5, _6, &denoising);