Project3: Xitong Zheng

CMakeLists.txt

@@ -32,7 +32,7 @@ set(CUDA_VERBOSE_BUILD ON)
 
 if(WIN32)
     # Set up include and lib paths
-    set(CUDA_HOST_COMPILER ${CMAKE_CXX_COMPILER} CACHE FILEPATH "Host side compiler used by NVCC" FORCE)
+    set(CUDA_HOST_COMPILER ${CMAKE_CXX_COMPILER} CACHE FILEPATH "Host side compiler used by NVCC" FORCE) 
 endif(WIN32)
 ########################################
 
@@ -114,8 +114,24 @@ source_group(Sources FILES ${sources})
 #add_subdirectory(src/ImGui)
 #add_subdirectory(stream_compaction)  # TODO: uncomment if using your stream compaction
 
+include_directories(./external/include)
+link_directories(./external/lib)
+
+
 cuda_add_executable(${CMAKE_PROJECT_NAME} ${sources} ${headers})
 target_link_libraries(${CMAKE_PROJECT_NAME}
     ${LIBRARIES}
     #stream_compaction  # TODO: uncomment if using your stream compaction
+	OpenImageDenoise
     )
+
+if (WIN32)
+    # Copy all DLL files from OIDN's bin directory to your executable's directory
+    file(GLOB OIDN_DLLS "${CMAKE_SOURCE_DIR}/external/bin/*.dll")
+
+    foreach(OIDN_DLL IN LISTS OIDN_DLLS)
+        add_custom_command(TARGET ${CMAKE_PROJECT_NAME} POST_BUILD
+                          COMMAND ${CMAKE_COMMAND} -E copy
+                          ${OIDN_DLL}  ${CMAKE_BINARY_DIR})
+    endforeach()
+endif()

README.md

@@ -1,13 +1,118 @@
 CUDA Path Tracer
-================
+==================================
 
 **University of Pennsylvania, CIS 565: GPU Programming and Architecture, Project 3**
 
-* (TODO) YOUR NAME HERE
-* Tested on: (TODO) Windows 22, i7-2222 @ 2.22GHz 22GB, GTX 222 222MB (Moore 2222 Lab)
+* Xitong Zheng
+  * [LinkedIn](https://www.linkedin.com/in/xitong-zheng-5b6543205/), [Instagram](https://www.instagram.com/simonz_zheng/), etc.
+* Tested on: Windows 11, i7-12700k 32GB, GTX 4090 24GB
 
-### (TODO: Your README)
+## Brief Intro
+This is a CUDA-based path tracer capable of rendering globally-illuminated images very quickly. \
+Why cuda? Because cuda is natively good at paralling rays and shading. To utiliza more parallism and improve occupancy, paths/rays are further divided into small path segments and processed in parallel. More performance improvement tricks are included in later part.
 
-*DO NOT* leave the README to the last minute! It is a crucial part of the
-project, and we will not be able to grade you without a good README.
+### Details
 
+#### Ideal Diffuse surfaces and Perfectly specular-reflective (mirrored) surfaces
+
+![](./img/image1.png)
+
+#### Path continuation/termination using Stream Compaction
+
+To avoid wasting time on paths that have already terminated, stream compaction is used to remove the terminated paths from the path pool. I use `thrust::stable_partition` to do this, which gives me an easy calculation of the still in-flight paths.
+
+#### Sort path based on material type
+
+Since path are randomly terminated for different reasons. Some hit light sources and other reach the maximum bounce limit. The memory access is not coalesced. To improve the performance, I sort the path based on the material type. This allows path rearrange in the blocks of materials and implys their status of keep bounceing or not. This should make the memory access pattern more coalesced and theoritally improve the performance.
+
+#### Toggleable cache first bounce intersections
+I cache the first bounce intersection for each path. This is because the first bounce intersection is the same for all the paths. So it is a waste of time to calculate it for each iteration.
+
+#### Refraction with Frensel effects using Schlick's approximation
+
+![](./img/refraction.png)
+
+#### Anti-aliasing
+![](./img/withantiliasing.png)
+
+With 4x anti-aliasing, the image is much smoother than the one without anti-aliasing. The image without anti-aliasing is shown below.
+
+![](./img/withoutantiliasing.png)
+#### Depth of field
+| depth of field with small aperture      | depth of field with big aperture      |
+|:-------------------------------------------:|:-------------------------------------------:|
+| ![](./img/dof_smallaperture.png) | ![](./img/dof_bigaperture.png)|
+
+
+
+#### Direct light
+
+| With direct light      | Without direct light      |
+|:-------------------------------------------:|:-------------------------------------------:|
+| ![](./img/withdirectlight.png) | ![](./img/withoutdirectlight.png)|
+
+With direct light, the image is brighter than the one without direct light.
+
+#### ObJ Mesh loading with bounding volume intersection culling
+
+I use `tinyObj` third-party loading code to bring the data to load the obj mesh. An Obj file can contains vertices, normals, texture coordinates, and faces in the following format:
+
+```
+v x y z
+
+The vertex command, this specifies a vertex by its three coordinates. 
+
+vt u v [w]
+
+The vertex texture command specifies the UV (and optionally W) mapping. 
+
+vn x y z
+
+The vertex normal command specifies a normal vector. 
+
+f v1[/vt1][/vn1] v2[/vt2][/vn2] v3[/vt3][/vn3] ...
+
+The face command specifies a polygon made from the verticies listed.
+```
+
+Here I use the vertices, vertex normal and faces to load the mesh and store them into related data structure. The intersection function used for mesh is `glm::intersectRayTriangle`.
+
+I also use bounding volume intersection culling to avoid unnecessary intersection calculation so as to reduce the number of rays that have to be checked against the entire mesh by first checking rays against a volume that completelt bounds the mesh. The bounding box is calculated by the min and max of the vertices of the mesh and can be created when the mesh is loaded. The intersection function used for bounding box is `glm::intersectRayAABB`. 
+
+
+![](./img/gear1.png)
+
+![](./img/vase.png)
+
+![](./img/vase1.png)
+#### Open Image AI denoiser
+
+As you can see that the above images are still noisy and not clear enough. And if the scene is complex and consists thousands of points and faces, you will end up into a situation that noise is eliminated slowly even if rendering for a relative long time. It can be anticipated that the marginal benefit of reducing noise use native monte carlo path tracing will be less and less.
+To improve the quality of the image, I use open image AI denoiser to denoise the image. The results are pretty good even if the spp is low. The result is shown below.
+![](./img/OIMD.png)
+
+
+The above two pictures are basically the same scene and you can easily find the outcome applied with open image AI denoiser is much clearear than the one without denoiser.
+
+Here I use first hit intersection to act as the albedo and normal filter of the pixel. Multiple filters have been added to the denoiser to improve the quality of the denoised image. The filters are listed below.
+
+- albedo filter
+- normal filter
+- basic(color/beauty) filter 
+
+Toggle `IMAGE_DENOISE` in the `main.cpp` to enable/disable the Open Image AI denoiser.
+
+#### Performance Analysis
+
+Stream compaction vs Non-stream compaction
+
+I use the gear scene to test the performance of stream compaction. 
+
+- bounding box intersection culling
+`#define BOUND_BOX 1` in the intersections.h to toggle the bounding box intersection culling. 
+The result is shown below.
+
+![](./img/performance1.png)
+
+#### Reference Links
+1. https://www.cs.cmu.edu/~mbz/personal/graphics/obj.html

external/include/OpenImageDenoise/config.h

@@ -0,0 +1,74 @@
+// Copyright 2018 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#define OIDN_VERSION_MAJOR 2
+#define OIDN_VERSION_MINOR 0
+#define OIDN_VERSION_PATCH 1
+#define OIDN_VERSION 20001
+#define OIDN_VERSION_STRING "2.0.1"
+
+/* #undef OIDN_API_NAMESPACE */
+
+#if defined(OIDN_API_NAMESPACE)
+  #define OIDN_API_NAMESPACE_BEGIN namespace  {
+  #define OIDN_API_NAMESPACE_END }
+  #define OIDN_API_NAMESPACE_USING using namespace ;
+  #define OIDN_API_EXTERN_C
+  #define OIDN_NAMESPACE ::oidn
+  #define OIDN_NAMESPACE_C _oidn
+  #define OIDN_NAMESPACE_BEGIN namespace  { namespace oidn {
+  #define OIDN_NAMESPACE_END }}
+#else
+  #define OIDN_API_NAMESPACE_BEGIN
+  #define OIDN_API_NAMESPACE_END
+  #define OIDN_API_NAMESPACE_USING
+  #if defined(__cplusplus)
+    #define OIDN_API_EXTERN_C extern "C"
+  #else
+    #define OIDN_API_EXTERN_C
+  #endif
+  #define OIDN_NAMESPACE oidn
+  #define OIDN_NAMESPACE_C oidn
+  #define OIDN_NAMESPACE_BEGIN namespace oidn {
+  #define OIDN_NAMESPACE_END }
+#endif
+
+#define OIDN_NAMESPACE_USING using namespace OIDN_NAMESPACE;
+
+#if defined(_WIN32)
+  #define OIDN_API_IMPORT OIDN_API_EXTERN_C __declspec(dllimport)
+  #define OIDN_API_EXPORT OIDN_API_EXTERN_C __declspec(dllexport)
+#else
+  #define OIDN_API_IMPORT OIDN_API_EXTERN_C
+  #define OIDN_API_EXPORT OIDN_API_EXTERN_C __attribute__((visibility ("default")))
+#endif
+
+#if defined(OpenImageDenoise_EXPORTS)
+  #define OIDN_API OIDN_API_EXPORT
+#else
+  #define OIDN_API OIDN_API_IMPORT
+#endif
+
+#if defined(_WIN32)
+  #define OIDN_DEPRECATED(msg) __declspec(deprecated(msg))
+#else
+  #define OIDN_DEPRECATED(msg) __attribute__((deprecated(msg)))
+#endif
+
+#if !defined(OIDN_DEVICE_CPU)
+  #define OIDN_DEVICE_CPU
+#endif
+#if !defined(OIDN_DEVICE_SYCL)
+  #define OIDN_DEVICE_SYCL
+#endif
+#if !defined(OIDN_DEVICE_CUDA)
+  #define OIDN_DEVICE_CUDA
+#endif
+#if !defined(OIDN_DEVICE_HIP)
+  #define OIDN_DEVICE_HIP
+#endif
+
+#define OIDN_FILTER_RT
+#define OIDN_FILTER_RTLIGHTMAP