diff --git a/CMakeLists.txt b/CMakeLists.txt
index 62c0e59..9f7c3c6 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -91,11 +91,13 @@ list(SORT sources)
source_group(Headers FILES ${headers})
source_group(Sources FILES ${sources})
-
-#add_subdirectory(stream_compaction) # TODO: uncomment if using your stream compaction
+include_directories(.)
+add_subdirectory(stream_compaction)
+add_subdirectory(tiny_obj)
cuda_add_executable(${CMAKE_PROJECT_NAME} ${sources} ${headers})
target_link_libraries(${CMAKE_PROJECT_NAME}
${LIBRARIES}
- #stream_compaction # TODO: uncomment if using your stream compaction
+ stream_compaction
+ tiny_obj
)
diff --git a/README.md b/README.md
index 110697c..3229e91 100644
--- a/README.md
+++ b/README.md
@@ -3,11 +3,111 @@ 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)
+* Disha Jindal: [Linkedin](https://www.linkedin.com/in/disha-jindal/)
+* Tested on: Windows 10 Education, Intel(R) Core(TM) i7-6700 CPU @ 3.40GHz 16GB, NVIDIA Quadro P1000 @ 4GB (Moore 100B Lab)
+## Path Tracer
+
-### (TODO: Your README)
+
-*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.
+## Overview
+This is an implementation of CUDA-based path tracer capable of rendering globally-illuminated images very quickly. Path tracing is a computer graphics Monte Carlo method of rendering images so that we can achieve good results with tracing a finite number out of the infinite space of rays.
+### Contents
+* `scenes/` Example scene description files
+* `img/` Renders of example scene description files
+* `external/` Includes and static libraries for 3rd party libraries
+* `src/` C++/CUDA source files
+ - `main.cpp` : Setup and keyboard control logic
+ - `pathtrace.cu` : Driver class which takes care of casting rays into the scene, testing for intersections, shading, graphics and performance optimizations and terminating a ray either after bouncing 8 times or reaching an emissive source
+ - `interactions.cu` : Simulates coloring and scattering of reflective, diffusing and refractive surfaces
+ - `intersections.cu` : Handles box, sphere, and mesh intersections
+ - `utilities.h` : Contains some utility functions and following flags to togge the features:
+ ```
+ #define COMPACT_RAYS [0,1]
+ #define CACHE_FIRST_BOUNCE [0,1]
+ #define MATERIAL_BASED_SORT [0,1]
+ #define ANTI_ALIASING [0,1]
+ #define MOTION_BLUR [0,1]
+ ```
+
+### Controls
+ * Esc to save an image and exit.
+ * S to save an image. Watch the console for the output filename.
+ * Space to re-center the camera at the original scene lookAt point
+ * left mouse button to rotate the camera
+ * right mouse button on the vertical axis to zoom in/out
+ * middle mouse button to move the LOOKAT point in the scene's X/Z plane
+
+## Features Implemented
+ * **Graphics**
+ - [x] Shaders
+ * Ideal Diffusion
+ * Perfect Reflection
+ * Refraction with fresnel effects [1/2 Additional Feature]
+ - [x] Antialiasing [1/2 Additional Feature]
+ - [x] Motion Blur [2/2 Additional Feature]
+ - [x] 3D Object Mesh Loading and Renderning [1/2 Extra Credit]
+ * **Optimizations**
+ - [x] Work-efficient shared memory based Stream Compaction [2/2 Extra Credit]
+ - [x] Contiguous rays by material type
+ - [x] Cache First Bounce
+
+### Ideal Diffusion
+A ray after striking with a material is either reflected, refracted or diffused depending upon the material properties of the object. Diffusion is implemented using Bidirectional Scattering Distribution Function.
+
+
+### Perfect Reflection
+In case of perfectly reflective surface, the new ray is calculated using `glm::reflect` function.
+
+
+### Refraction with fresnel effects
+Refraction is calculated using Snell's law and I have used `glm::refract` function to do this. But since most materials are not perfectly sepcular, have implmented fresnel effects using **Schlick's approximation**. Fresnel equations give the proportion of reflected and refracted light and then a random number from 0 to 1 is calculated to choose between specular reflection and refraction.
+
+
+### Antialiasing
+Antialiasing is a technique to diminish the jaggies/stairstep-like lines and smoothen them. This is implemented using a very simple trick that is by jittering the pixel's location. The idea is to subdivide the pixel into subpixels and choose a random supixel each time rather than always looking at the center to of the pixel. Accumulating the effect across multiple iterations, the intensity value of the pixel is the average of all these samples and creates a more continuos effect.
+
+
+### Motion Blur
+Motion blur is another technique which leverages this averaging effect of this implementation. To implement this, the object is moved slighlty between each iteration and the averaging of such multiple shots creats the effect of motion.
+
+
+
+### 3D Object Modeling
+Loading 3D models (Reference: https://free3d.com/) using [tinyObj](http://syoyo.github.io/tinyobjloader/) and then checking triangle intersection using `glm::intersectRayTriangle`.
+
+
+
+
+
+## Optimizations
+### Stream Compaction
+A lot of rays die after a few iterations by either merging into light or the ones which do not intersect with any object. So, we can use stream compaction to limit the number of rays we are tracing and the number of threads launched at each iteration. I am using my Work-efficient stream compaction implementation across multiple blocks which uses shared memory for performance.
+
+#### Performance impact of stream compaction
+Following plot shows the average time per depth with and without stream compaction. Stream compaction took around 3.8 ms whereas it took 4s without it. These are the results with 8 bounces and so the performance would increase even further with more bounces and more complex scenes.
+
+
+
+#### Number of live/unterminated rays at each iteration
+Following plots shows the number of unterminated rays at each depth. Yellow bars correspond to an open scene and the red bars show corresponding closed scene with additional left and right walls. We can see that the numbe rof live rays drop at a very fast pace in the open scene compared to the closed one.
+
+
+
+### Contiguous rays by material type
+The shader implementation depends on the material with which the ray has intersected. So, If one warp has rays intersecting with different materials, it would lead to warp divergence and only the threads with one material could run at one time making it sequential in the number of materials. We can avoid this performance bottleneck by sorting the rays according to the material they are intersecting with so that rays interacting with the same material are contiguous in memory before shading and warp divergence is reduced.
+
+Following plot shows the average time per iteration with and without sorting tha paths according to the material type. There is a huge performance drop due to this. One potential reason for this is the number of materials (6 in this case) used to create the scene .Another reason is the sorting overhead. Probably the gain due to less warp divergence is not sufficient to make up for the sorting overhead. It might help in case we have a huge number of materials.
+
+
+
+### Cache First Bounce
+One first step of generating rays and finding itersection for the first bounce is same across all iterations with an exception while we are using anti aliasing. So, we could to an optimization by saving the first after the first iteration and reuse it rather than re doing it every time.
+
+Following plot shows average time per iteration with and without using cache. It took around 32 ms for iteration without using cache and 29 ms with cache. These number are calculated with an average across 10 iterations. The gap would increase with the complexity of the scene specifically the number of objects.
+
+
+## Bloopers
+Following are some of the bloopers. First one was caused when I used an offset of 0.00001f instead of 0.0001f. The second was when I gave the reverse of eta to the refract function instead of eta.
+
diff --git a/external/include/json.hpp b/external/include/json.hpp
new file mode 100644
index 0000000..9912bc7
--- /dev/null
+++ b/external/include/json.hpp
@@ -0,0 +1,20405 @@
+/*
+ __ _____ _____ _____
+ __| | __| | | | JSON for Modern C++
+| | |__ | | | | | | version 3.5.0
+|_____|_____|_____|_|___| https://github.com/nlohmann/json
+
+Licensed under the MIT License .
+SPDX-License-Identifier: MIT
+Copyright (c) 2013-2018 Niels Lohmann .
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+*/
+
+#ifndef NLOHMANN_JSON_HPP
+#define NLOHMANN_JSON_HPP
+
+#define NLOHMANN_JSON_VERSION_MAJOR 3
+#define NLOHMANN_JSON_VERSION_MINOR 5
+#define NLOHMANN_JSON_VERSION_PATCH 0
+
+#include // all_of, find, for_each
+#include // assert
+#include // and, not, or
+#include // nullptr_t, ptrdiff_t, size_t
+#include // hash, less
+#include // initializer_list
+#include // istream, ostream
+#include // random_access_iterator_tag
+#include // accumulate
+#include // string, stoi, to_string
+#include // declval, forward, move, pair, swap
+
+// #include
+#ifndef NLOHMANN_JSON_FWD_HPP
+#define NLOHMANN_JSON_FWD_HPP
+
+#include // int64_t, uint64_t
+#include