CIS565-Fall-2019 · karvesaket · Sep 25, 2019 · Sep 25, 2019 · Sep 25, 2019 · Sep 26, 2019
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -92,10 +92,14 @@ list(SORT sources)
 source_group(Headers FILES ${headers})
 source_group(Sources FILES ${sources})
 
-#add_subdirectory(stream_compaction)  # TODO: uncomment if using your stream compaction
+include_directories(.)
+add_subdirectory(stream_compaction)  # TODO: uncomment if using your stream compaction
+add_subdirectory(tiny_obj)  # TODO: uncomment if using your stream compaction
+
 
 cuda_add_executable(${CMAKE_PROJECT_NAME} ${sources} ${headers})
 target_link_libraries(${CMAKE_PROJECT_NAME}
     ${LIBRARIES}
-    #stream_compaction  # TODO: uncomment if using your stream compaction
+    stream_compaction  # TODO: uncomment if using your stream compaction
+	tiny_obj
     )
diff --git a/README.md b/README.md
@@ -3,11 +3,167 @@ 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)
+* Saket Karve
+  * [LinkedIn](https://www.linkedin.com/in/saket-karve-43930511b/), [twitter](), etc.
+* Tested on:  Windows 10 Education, Intel(R) Core(TM) i7-6700 CPU @ 3.40GHz 16GB, NVIDIA Quadro P1000 @ 4GB (Moore 100B Lab)
 
-### (TODO: Your README)
+### Highlights
 
-*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.
+![](img/gold_cubes.png)
 
+![](img/gold.png)
+
+![](img/complex_without_cube_yellow.png)
+
+![](img/cornell_penn.png)
+
+### Features implemented
+
+- Visual artefacts
+  - Shading different materials
+    - Ideal diffuse and shading
+    - Perfect Specular reflection
+    - Refraction with Fresnel effects and total internal reflection \[EXTRA CREDIT\]
+  - Stochastic Anti-aliasing \[EXTRA FEATURE\]
+  - Motion Blur \[EXTRA FEATURE\]
+  - Arbitrary mesh loading and rendering (OBJ) \[EXTRA FEATURE\]
+- Performance improvements
+  - Path termination using Stream Compaction
+  - Cache first bounce
+  - Sort by materials
+  - Stream Compaction using shared memory \[EXTRA FEATURE\]
+
+### Shading different materials
+
+![](img/cornell_multiple.png)
+
+#### Ideal Diffuse
+
+When rays fall on an object with ideal diffuse, the ray is scattered randomnly sampled from a uniform distribution in any direction within the hemisphere centered at the point of incidence. This gives a matt finish to the object.
+
+![](img/cornell_diffuse.png)
+
+#### Perfect Specular reflection
+
+When rays fall on an object with perfectly specular material, it always (with 100% probability) reflects at an angle equal to the incidence angle on the other side of the normal. This gives a shiny mirror like finish to the object. Reflectons of objects around can be seen. Since it does not allow any light to pass through, we can see a shadow on the side which is away from light.
+
+![](img/cornell_perfect_reflection_1.png)
+
+#### Refraction
+
+When a ray falls on a refractive surface, it penetrates inside the object making an angle with the normal determined by the refractive index of the material the rays comes from and that of the object. This angle is determined by Snell's law. 
+
+A perfectly refractive object is rendered as below.
+
+![](img/cornell_refr_15_without_fresnel.png)
+
+However, objects are not perfectly refractive. Some proportion of the rays are reflected depending on the refractive indices of the materials and the angle of incidence of the ray. The reflectance coefficient (proportion of rays which refract) is calculated by Fresnel's law. Also, for rays coming from a material with higher refractive index to one with a lower refractive index, some rays reflect perfectly if the angle of incidence is more than a perticuar value (called critical angle). This is called total internal reflection. 
+
+The render of an object with refraction wil be as follows.
+
+![](img/cornell_refraction_15.png)
+
+Render of a sphere with refractive index = 2.41 (diamond)
+
+![](img/cornell_refraction_241.png)
+
+Render of sphere with 50% reflection and 50% chance of refraction (with Fresnel's effect)
+
+![](img/cornell_half_refl_half_refr.png)
+
+### Anti-Aliasing
+
+Depending on the resolution, when images are rendered, the pixels show a stair-step like lines near the edges of objects. This can be reduced by approximating the pixel values around that pixel. This makes the edge more smooth. For thie path tracer, this is implemented by adding some random noise (jitter) to the pixel value when generating rays from the camera. This means, we shoot a ray to a point in an area around the actual pixel randomly sampling from a unifrom distribution. At every iteration, the ray strikes at a different point around the pixel (stochastically) and thus across iterations, creates an average effect at each pixel.
+
+The render with and without antialiasing can be seen in the following figure. These renders are obtained after running the path tracer for 2000 iterations for both cases.
+
+| Without Antialiasing | With visualization |
+| ---------------------|---------------------- |
+| ![](img/cornell_without_AA_2000.png) | ![](img/cornell_with_AA_2000.png)                 |
+
+The zoomed in version of the above images are as follows. This shows how the render is jittery without antialiasing more clearly.
+
+| Without Antialiasing | With Antialiasing |
+| ---------------------|---------------------- |
+| ![](img/without_AA_zoom.PNG) | ![](img/with_AA_zoom.PNG)                 |
+
+### Motion Blur
+
+Motion blur shows the effect of an object in motion which results when the image being recorded changes due to long exposure to the camera. The following renders are obtained for an object with ideal diffuse and a refractive object with motion blur.
+
+![](img/cornell_modtion_blur_diffuse.png)
+
+### Arbitrary mesh loading and rendering
+
+Different objects are loaded from their OBJ files using the TinyObj library functions. Objects are loaded as a list of triangles represented by their vertices. To render these objects, the intersection of the ray is computed with all the triangles part of the object (mesh) to check if it intersects with the object. If it does, the closest point of intersection among all the triangles is determined.
+
+Different objects were loaded in the scene.
+
+![](img/droid.png)
+
+### Path termination using stream compaction
+
+When rays are either generated from the camera or bounced off some object do not hit any object in the scene the rays are terminated. This means these rays do not bounce further. Since our implementation of path tracing is per ray i.e. we launch threads per ray (rather than per pixel), the threads for such terminated rays are simply idling. So, if we compact the array of rays after every bounce we can save a lot of GPU resources by launching only as many thread as there are active rays.
+
+The following graph shows the remaining active rays after certain number of iterations.
+
+![](img/sc_rem_bounces.PNG)
+
+Time taken to complete 1 iteration (Measured across 10 iterations)
+
+- Without stream compaction = 0.0343 sec
+- With stream compaction = 0.0735 sec
+
+Stream compaction has a greater effect when the scene is open as opposed to cornell which is a closed scene. With an open scene, the time taken for 1000 iterations with stream compaction is,
+
+- Closed scene = 0.0735 sec
+- Open scene = 0.0303 sec
+
+
+### Cache First Bounce
+
+Since the first ray which is generated from the camera hits the same point, the first bounce will be the same for all iterations. Diffused materials scatter the ray in some random direction with some probability, but this caching does not cause any difference in the render even if we always use the same first bounce. So, I save the first bounce for each ray and load it directly from the cached version at each iteration.
+
+Time taken to complete 1 iteration (Measured across 10 iterations)
+
+- Without caching  = 0.0761 sec
+- With caching = 0.0741 sec
+
+### Sort by materials
+
+Sorting rays by the materials it hits allows the threads in a single warp to have more chances of hitting the same material. This reduces warp divergence and is thus expected to improve performance. The imporvement in performance is not visible for simple scenes with less objects. For complex scenes, the image rendering takes lesser time when sorted.
+
+Time taken to complete 1 iteration (Measured across 10 iterations)
+
+- Without Sorting = 0.0770 sec
+- With Sorting = 0.214 sec
+
+Since the scene had very few materials, sorting actually did not help as it should. However, if the scene is quite complex, then material sorting should make it much more efficient per iteration.
+
+### Stream Compaction with shared memory
+
+I implemented stream compaction using shared memory to improve stream compaction's performance. The stream compaction analysis shown above is done using Thrust's ```thrust::partition()``` function. The work efficient stream compaction implemented in the [previous project]() has a lot of global memory accesses for the up-sweep and down-sweep. This creates a bottleneck as global memory is slow to access and thus reduces efficiency. Using shared memory helps reduce the global memory access by loading required data to shared memory and then doing the required computations. However, since shared memory is limited and threads can not be synced across blocks, so this implementation is not trivial. 
+
+I implemented the stream compaction with shared memory in project 2's repository and it resulted in the following improvement in time for stream compaction.
+
+![](img/sc_plot.PNG)
+
+The above graph shows how shared memory takes lesser time with increasing array size for stream compaction.
+
+When adding stream compaction with shared memory to the Path Tracer, I was getting some Out of Memory issues for images with 800 x 800 resolution. The path tracer works perfectly with lesser resolution of upto 400 x 400
+
+### Bloopers
+
+Some of the really interesting bloopers I got while implementing this project.
+
+- When the random seed was not unique for every bounce of the ray, some wierd banding was observed.
+
+![](img/blooper_banding_seed.png)
+
+- When the index of refraction was reciprocal of what is should be.
+
+![](img/blooper_donut_incorrect_ior.png)
+
+- When a very small noise was added to the origin of the scattered ray.
+
+![](img/blooper_full_white.png)
diff --git a/img/India1.png b/img/India1.png
diff --git a/img/India_reflective.png b/img/India_reflective.png
diff --git a/img/blooper_banding_seed.png b/img/blooper_banding_seed.png
diff --git a/img/blooper_donut_incorrect_ior.png b/img/blooper_donut_incorrect_ior.png
diff --git a/img/blooper_full_white.png b/img/blooper_full_white.png
diff --git a/img/complex_rotated_cube_white.png b/img/complex_rotated_cube_white.png
diff --git a/img/complex_without_cube_white.png b/img/complex_without_cube_white.png
diff --git a/img/complex_without_cube_yellow.png b/img/complex_without_cube_yellow.png
diff --git a/img/cornell_diffuse.png b/img/cornell_diffuse.png
diff --git a/img/cornell_half_refl_half_refr.png b/img/cornell_half_refl_half_refr.png
diff --git a/img/cornell_modtion_blur_diffuse.png b/img/cornell_modtion_blur_diffuse.png
diff --git a/img/cornell_mption_blur_refr.png b/img/cornell_mption_blur_refr.png
diff --git a/img/cornell_multiple.png b/img/cornell_multiple.png
diff --git a/img/cornell_penn.png b/img/cornell_penn.png
diff --git a/img/cornell_perfect_reflection_1.png b/img/cornell_perfect_reflection_1.png
diff --git a/img/cornell_refr_15_without_fresnel.png b/img/cornell_refr_15_without_fresnel.png
diff --git a/img/cornell_refraction_133.png b/img/cornell_refraction_133.png
diff --git a/img/cornell_refraction_15.png b/img/cornell_refraction_15.png
diff --git a/img/cornell_refraction_241.png b/img/cornell_refraction_241.png
diff --git a/img/cornell_with_AA_2000.png b/img/cornell_with_AA_2000.png
diff --git a/img/cornell_without_AA_2000.png b/img/cornell_without_AA_2000.png
diff --git a/img/droid.png b/img/droid.png
diff --git a/img/gold.png b/img/gold.png
diff --git a/img/gold_cubes.png b/img/gold_cubes.png
diff --git a/img/sc_plot.PNG b/img/sc_plot.PNG
diff --git a/img/sc_rem_bounces.PNG b/img/sc_rem_bounces.PNG
diff --git a/img/stream_compaction.PNG b/img/stream_compaction.PNG
diff --git a/img/with_AA_zoom.PNG b/img/with_AA_zoom.PNG
diff --git a/img/without_AA_zoom.PNG b/img/without_AA_zoom.PNG
diff --git a/objects/BlackAndRedFloatingRobot.mtl b/objects/BlackAndRedFloatingRobot.mtl
@@ -0,0 +1,20 @@
+# Blender MTL File: 'BlackAndRedFloatingRobot.blend'
+# Material Count: 2
+
+newmtl Material.001
+Ns 96.078431
+Ka 0.000000 0.000000 0.000000
+Kd 0.000000 0.000000 0.000000
+Ks 0.500000 0.500000 0.500000
+Ni 1.000000
+d 1.000000
+illum 2
+
+newmtl Material.002
+Ns 96.078431
+Ka 0.000000 0.000000 0.000000
+Kd 1.000000 0.028567 0.000000
+Ks 0.500000 0.500000 0.500000
+Ni 1.000000
+d 1.000000
+illum 2
diff --git a/objects/BrandenburgGate.mtl b/objects/BrandenburgGate.mtl
@@ -0,0 +1,71 @@
+newmtl Preview
+Ns 50
+Ka 0 0 0
+Kd 0.925138 0.781762 0.639699
+Ks 0.0915311 0.0604877 0.189959
+newmtl 09_-_BronzeStatues
+Ns 10
+Ka 0 0 0
+Kd 0.8 0.8 0.8
+Ks 0 0 0
+map_Kd BronzeStatues1.jpg
+newmtl 02_-_Marble
+Ns 10
+Ka 0 0 0
+Kd 0.8 0.8 0.8
+Ks 0 0 0
+map_Kd MarbleColumns2.jpg
+newmtl 04_-_BronzeRoof
+Ns 10
+Ka 0 0 0
+Kd 0.8 0.8 0.8
+Ks 0 0 0
+map_Kd BronzeRoof1.jpg
+newmtl 05_-_Fregio
+Ns 10
+Ka 0 0 0
+Kd 0.8 0.8 0.8
+Ks 0 0 0
+map_Kd Fregio1.jpg
+newmtl 05_-_MarbleColumns
+Ns 10
+Ka 0 0 0
+Kd 0.8 0.8 0.8
+Ks 0 0 0
+map_Kd MarbleColumns2.jpg
+newmtl 03_-_Wall
+Ns 10
+Ka 0 0 0
+Kd 0.8 0.8 0.8
+Ks 0 0 0
+map_Kd WhiteWall1.jpg
+newmtl 08_-_DarkerWall
+Ns 10
+Ka 0 0 0
+Kd 0.8 0.8 0.8
+Ks 0 0 0
+map_Kd DarkerWall1.jpg
+newmtl 01_-_Columns
+Ns 10
+Ka 0 0 0
+Kd 0.8 0.8 0.8
+Ks 0 0 0
+map_Kd Columns1.jpg
+newmtl 06_-_SideGliph
+Ns 10
+Ka 0 0 0
+Kd 0.8 0.8 0.8
+Ks 0 0 0
+map_Kd SideGliph1.jpg
+newmtl 01_-_TempleFront
+Ns 10
+Ka 0 0 0
+Kd 0.8 0.8 0.8
+Ks 0 0 0
+map_Kd TempleFront1.jpg
+newmtl 07_-_SideBuilding
+Ns 10
+Ka 0 0 0
+Kd 0.8 0.8 0.8
+Ks 0 0 0
+map_Kd WindowsBuilding1.jpg
diff --git a/objects/Pikachu MTL.mtl b/objects/Pikachu MTL.mtl
@@ -0,0 +1,46 @@
+# Blender MTL File: 'Pikachu For Sketchfab.blend'
+# Material Count: 4
+
+newmtl Cheek
+Ns 96.078431
+Ka 1.000000 1.000000 1.000000
+Kd 0.229953 0.000486 0.008264
+Ks 0.500000 0.500000 0.500000
+Ke 0.000000 0.000000 0.000000
+Ni 1.000000
+d 1.000000
+illum 2
+map_Kd C:\Blender EXE\Character\Pokedex3D_art_1.jpg
+
+newmtl Eye,_Ear
+Ns 96.078431
+Ka 1.000000 1.000000 1.000000
+Kd 0.000000 0.000000 0.000000
+Ks 0.500000 0.500000 0.500000
+Ke 0.000000 0.000000 0.000000
+Ni 1.000000
+d 1.000000
+illum 2
+map_Kd C:\Blender EXE\Character\Pokedex3D_art_1.jpg
+
+newmtl Eye_02
+Ns 96.078431
+Ka 1.000000 1.000000 1.000000
+Kd 0.640000 0.640000 0.640000
+Ks 0.500000 0.500000 0.500000
+Ke 0.000000 0.000000 0.000000
+Ni 1.000000
+d 1.000000
+illum 2
+map_Kd C:\Blender EXE\Character\Pokedex3D_art_1.jpg
+
+newmtl Skin
+Ns 96.078431
+Ka 1.000000 1.000000 1.000000
+Kd 0.717674 0.513855 0.002238
+Ks 0.500000 0.500000 0.500000
+Ke 0.000000 0.000000 0.000000
+Ni 1.000000
+d 1.000000
+illum 2
+map_Kd C:\Blender EXE\Character\Pokedex3D_art_1.jpg
diff --git a/objects/android.mtl b/objects/android.mtl
@@ -0,0 +1,20 @@
+# Blender MTL File: 'android.blend'
+# Material Count: 2
+
+newmtl android
+Ns 96.078431
+Ka 0.000000 0.000000 0.000000
+Kd 0.296990 0.451769 0.032732
+Ks 0.500000 0.500000 0.500000
+Ni 1.000000
+d 1.000000
+illum 2
+
+newmtl oko
+Ns 96.078431
+Ka 0.000000 0.000000 0.000000
+Kd 0.000000 0.000000 0.000000
+Ks 0.500000 0.500000 0.500000
+Ni 1.000000
+d 1.000000
+illum 2
diff --git a/objects/bb8.mtl b/objects/bb8.mtl
@@ -0,0 +1,46 @@
+# 3ds Max Wavefront OBJ Exporter v0.97b - (c)2007 guruware
+# File Created: 30.01.2016 20:42:29
+
+newmtl wire_153228153
+	Ns 32
+	d 1
+	Tr 0
+	Tf 1 1 1
+	illum 2
+	Ka 0.6000 0.8941 0.6000
+	Kd 0.6000 0.8941 0.6000
+	Ks 0.3500 0.3500 0.3500
+
+newmtl Material__61
+	Ns 10.0000
+	Ni 1.5000
+	d 1.0000
+	Tr 0.0000
+	Tf 1.0000 1.0000 1.0000 
+	illum 2
+	Ka 0.7176 0.6902 0.6588
+	Kd 0.7176 0.6902 0.6588
+	Ks 0.0000 0.0000 0.0000
+	Ke 0.0000 0.0000 0.0000
+
+newmtl Material__62
+	Ns 10.0000
+	Ni 1.5000
+	d 1.0000
+	Tr 0.0000
+	Tf 1.0000 1.0000 1.0000 
+	illum 2
+	Ka 0.0275 0.0275 0.0275
+	Kd 0.0275 0.0275 0.0275
+	Ks 0.0000 0.0000 0.0000
+	Ke 0.0000 0.0000 0.0000
+
+newmtl wire_028149177
+	Ns 32
+	d 1
+	Tr 0
+	Tf 1 1 1
+	illum 2
+	Ka 0.1098 0.5843 0.6941
+	Kd 0.1098 0.5843 0.6941
+	Ks 0.3500 0.3500 0.3500