Term Project

External libraries used:
- tinyxml2 (tinyxml2.h, tinyxml2.cpp are provided in the homework files.)
- lodepng (lodepng.h, lodepng.cpp are provided in the homework files.) 
- Eigen (Can be obtained from http://eigen.tuxfamily.org/) 
- jpeglib

Command to compile:
g++ -I /usr/include/eigen3/ -std=c++11 -O3 main.cpp RayTracer.cpp ray.cpp shape.cpp light.cpp tinyxml2.cpp lodepng.cpp BVH.cpp texture.cpp -o RayTracer -ljpeg

Computer specs:
Intel Core i7-2630QM CPU @ 2.00 GHz, 6 GB RAM

Running the ray tracer:
./RayTracer command should be fed by the relative path of the input file, e. g.,
./RayTracer demo.xml

Related articles:
1. http://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=4634627
2. https://cseweb.ucsd.edu/~ravir/papers/singlescat/scattering.pdf

Term project's topic was to choose a model to simulate participating media (e. g., fog, haze, smoke) and implement it in ray tracer. 


A very simple input file and executable for it is given in the ling below: 
https://drive.google.com/drive/folders/0ByiItrg43ou-OVNmcHdVN08xLWs?usp=sharing

Far-away objects should be seen less clear than near objects due to participating media.

There is one point light in the scene.

Anti-Aliasing: MSAA w/ 25 samples
Render Time: 92m 14s

In previous assignments, it was not possible to render a point light even it directly sees the camera (Because a point light has no volume). In the image left, a point light seen as if it had a volume thanks to the participating media's aerial glow effect.

There is one point light in the scene.

Anti-Aliasing: MSAA w/ 16 samples
Render Time: 113m 56s

Spot lights forms a light cone due to the participating media.

There is one spot light in the scene.

Render Time: 13m 4s

Note:
sigma_a = (0.1, 0.1, 0.1)
sigma_s = (0.3, 0.3, 0.3)

Spot light radiates only white light, also the surface absorbs and reflects all the spectrum equally. The color is due to fact that in this scene participating media scatters each color channel differently.

There is one spot light in the scene.

Render Time: 11m 27s

Note:
sigma_a = (0.07, 0.07, 0.07)
sigma_s = (0.1471, 0.2930, 0.3000)

Volumetric shadows with basic geometry.

There is one spot light in the scene.

Anti-Aliasing: MSAA w/ 36 samples
Render Time: 497m 58s

Volumetric shadows with complex geometry.

There is one spot light in the scene.

Anti-Aliasing: MSAA w/ 36 samples
Render Time: 650m 42s

Homework 8

External libraries used:
- tinyxml2 (tinyxml2.h, tinyxml2.cpp are provided in the homework files.)
- lodepng (lodepng.h, lodepng.cpp are provided in the homework files.) 
- Eigen (Can be obtained from http://eigen.tuxfamily.org/) 
- jpeglib

Command to compile:
g++ -I /usr/include/eigen3/ -std=c++11 -O3 main.cpp RayTracer.cpp ray.cpp shape.cpp light.cpp tinyxml2.cpp lodepng.cpp BVH.cpp texture.cpp -o RayTracer -ljpeg

Computer specs:
Intel Core i7-2630QM CPU @ 2.00 GHz, 6 GB RAM

Running the ray tracer:
./RayTracer command should be fed by the relative path of the input file, e. g.,
./RayTracer 795_input_set_06/skybox.xml

This week's task was to implement more realistic BRDFs (Bi-Directional Reflectance Distribution Function), specifically, Blinn-Phong, Phong, and Torrance-Sparrow BRDFs. 

File Name: brdf_blinnphong_modified.xml
BRDF: Modified Blinn-Phong (Unnormalized)

File Name: brdf_blinnphong_modified_normalized.xml
BRDF: Modified Blinn-Phong (Normalized)

File Name: brdf_blinnphong_original.xml
BRDF: Original Blinn-Phong 

File Name: brdf_phong_modified.xml
BRDF: Modified Phong (Unnormalized)

File Name: brdf_phong_modified_normalized.xml
BRDF: Modified Phong (Normalized)

File Name: brdf_phong_original.xml
BRDF: Original Phong

File Name: brdf_torrancesparrow.xml
BRDF: Torrance-Sparrow

File Name: killeroo_blinnphong_modified_normalized.xml
BRDF: Modified Blinn-Phong (Normalized)

Image just below is the close-up version. 

File Name: killeroo_torrancesparrow.xml
BRDF: Torrance-Sparrow

Image just below is the close-up version. 

Comments: At the left-bottom part of the close-up image, some green artifacts can be seen. I could not figure out which bug caused these artifacts yet. 

Homework 7

External libraries used:
- tinyxml2 (tinyxml2.h, tinyxml2.cpp are provided in the homework files.)
- lodepng (lodepng.h, lodepng.cpp are provided in the homework files.) 
- Eigen (Can be obtained from http://eigen.tuxfamily.org/) 
- jpeglib

Command to compile:
g++ -I /usr/include/eigen3/ -std=c++11 -O3 main.cpp RayTracer.cpp ray.cpp shape.cpp light.cpp tinyxml2.cpp lodepng.cpp BVH.cpp texture.cpp -o RayTracer -ljpeg

Computer specs:
Intel Core i7-2630QM CPU @ 2.00 GHz, 6 GB RAM

Running the ray tracer:
./RayTracer command should be fed by the relative path of the input file, e. g.,
./RayTracer 795_input_set_06/skybox.xml

This week's task was to implement bump mapping which creates a sense of depth in texture mapping. 

File Name: bump_mapping_basic.xml
Anti-Aliasing: MSAA with 36 sample per pixel

File Name: sphere_bump_nobump.xml
Anti-Aliasing: MSAA with 36 sample per pixel

File Name: killeroo_bump_walls.xml
Anti-Aliasing: MSAA with 16 sample per pixel
Render Time: 1 minute 51.583 seconds

File Name: bump_mapping_transformed.xml
Anti-Aliasing: MSAA with 36 sample per pixel
Comments: Scaling problem in spheres seems to be solved. However, cobblestone texture does not seems to rotated exactly in the outputs provided probably due to a small bug.

Homework 6

External libraries used:
- tinyxml2 (tinyxml2.h, tinyxml2.cpp are provided in the homework files.)
- lodepng (lodepng.h, lodepng.cpp are provided in the homework files.) 
- Eigen (Can be obtained from http://eigen.tuxfamily.org/) 
- jpeglib

Command to compile:
g++ -I /usr/include/eigen3/ -std=c++11 -O3 main.cpp RayTracer.cpp ray.cpp shape.cpp light.cpp tinyxml2.cpp lodepng.cpp BVH.cpp texture.cpp -o RayTracer -ljpeg

Computer specs:
Intel Core i7-2630QM CPU @ 2.00 GHz, 6 GB RAM

Running the ray tracer:
./RayTracer command should be fed by the relative path of the input file, e. g.,
./RayTracer 795_input_set_06/skybox.xml

This week's task was to implement texture mapping. 

File Name: sphere_texture_replace_nearest.xml
Interpolation Method: Nearest
Decal Mode: Replace kd

File Name: sphere_texture_replace_bilinear.xml
Interpolation Method: Bilinear
Decal Mode: Replace kd

File Name: sphere_texture_blend_bilinear.xml
Interpolation Method: Bilinear
Decal Mode: Blend kd

File Name: killeroo_diffuse_specular_texture.xml
Interpolation Method: Bilinear
Decal Mode: Blend kd
Appearance: Repeat

File Name: perlin_types.xml
Comments:
- Perlin noise examples. Smooth textures were generated in patch mode, textures with veins were generated in vein mode.
- There is also a directional light in the scene.

File Name: simple_texture.xml
Comments:
- Texture on the sphere in top-left was generated with nearest interpolation and replace kd modes.
- Texture on the sphere in bottom-left was generated with patch and replace kd modes.
- Texture on the sphere in bottom-right was generated with vein and replace kd modes.
- Textures on the square and the triangle were generated in nearest and replace kd modes.

File Name: skybox.xml

Anti-Aliasing: MSAA with 100 sample per pixel

Render time: 2 minutes 52.054 seconds

- Texture of the sky was generated in bilinear and replace all modes. 

- Texture of the floor was generated in bilinear and replace kd modes with repeated appearance. 

Other Comments and Highlights: 

- elliosoids_texture.xml could not be rendered because of the bug in scaling the spheres (i.e., ellipsoid shape cannot be obtained). 

- All in all, texture mapping seems to be working fine.

- There seems to be bugs related with the tasks of the previous weeks (such as transformations). 

Homework 5

External libraries used:
- tinyxml2 (tinyxml2.h, tinyxml2.cpp are provided in the homework files.)
- lodepng (lodepng.h, lodepng.cpp are provided in the homework files.) 
- Eigen (Can be obtained from http://eigen.tuxfamily.org/) 

Command to compile:
g++ -I /usr/include/eigen3/ -std=c++11 -O3 main.cpp RayTracer.cpp Ray.cpp shape.cpp light.cpp tinyxml2.cpp lodepng.cpp BVH.cpp -o RayTracer

Computer specs:
Intel Core i7-2630QM CPU @ 2.00 GHz, 6 GB RAM

Running the ray tracer:
./RayTracer command should be fed by the relative path of the input file, e. g.,
./RayTracer 795_input_set_05/spheres_dof.xml

This week's task was to implement multi sampling anti aliasing (MSAA), spot light, area light, glossy reflection and depth of field.


File Name: dragon_spot_light.xml
Render Time: 16.789 seconds
Comments: No MSAA only spot light.

File Name: dragon_spot_light_msaa.xml
Number of Samples: 100
Render Time: 27 minutes 21.153 seconds
Comments: MSAA and spot light.

File Name: glass_plates_point.xml
Number of Samples: 36
Render Time: 4 minutes 33.520 seconds
Comments: MSAA and point light.















File Name: glass_plates_area.xml
Number of Samples: 36
Render Time: 3 minutes 45.235 seconds
Comments: MSAA and area light.

File Name: metal_plates_area.xml
Number of Samples: 36
Render Time: 3 minutes 39.345 seconds
Comments: MSAA and area light.

File Name: spheres_dof.xml
Number of Samples: 100
Render Time: 36.618 seconds
Comments: MSAA and point light plus finite aperture size which simulates a realistic camera lens.

Homework 4

External libraries used:
- tinyxml2 (tinyxml2.h, tinyxml2.cpp are provided in the homework files.)
- lodepng (lodepng.h, lodepng.cpp are provided in the homework files.) 
- Eigen (Can be obtained from http://eigen.tuxfamily.org/) 

Command to compile:
g++ -I /usr/include/eigen3/ -std=c++11 -O3 main.cpp RayTracer.cpp Vec3d.cpp Ray.cpp shape.cpp tinyxml2.cpp lodepng.cpp BVH.cpp -o RayTracer

Computer specs:
Intel Core i7-2630QM CPU @ 2.00 GHz, 6 GB RAM

Running the ray tracer:
./RayTracer command should be fed by the relative path of the input file, e. g.,
795_input_set_02/horse.xml

This week's task was to implement reflection and refraction, more specifically, implementation of Snell's Law, Schlick's approximation to Fresnel equations and Beer's Law. 

File Name: cornellbox_glass.xml
Render Time: 9.088 seconds

File Name: glass_plates.xml
Render Time: 12.981 seconds
Comments: Although refraction and transmission phenomena looks to be implemented correctly, the image looks as if we are looking at the scene at a slightly different angle than the original output. The problem seems to be related with the camera model implementation.

File Name: horse_and_glass_mug.xml
Render Time: ~1 minute 28 seconds

File Name: killeroo_glass.xml
Render Time: 44.333 seconds
Comments: When I run the .xml file provided as it is, the image looks like we are looking at the scene from just outside of the green wall (was discussed in the class.). To better understand the problem, I took all the box meshes out except the one under the dino (with mesh id's 3, 4, 5, 6, 7), and added the walls one by one.

File Name: killeroo_glass.xml
Render Time: ~45 seconds
Comments: Green wall added (mesh id 3).

File Name: killeroo_glass.xml
Render Time: ~45 seconds

Comments: Blue wall added (mesh id 4).

When the other walls are added, the image is created as if the camera outside the box. The problem seems to be related with triangle intersection routine (Especially when the triangle is at the back of the camera.) but I have not solved it yet. 

File Name: killeroo_half_mirror.xml
Render Time: 30.315 seconds
Comments: As in killeroo_glass.xml, meshes with id's 5, 6, and 7 are excluded.

File Name: killeroo_mirror.xml
Render Time: 28.262 seconds
Comments: As in killeroo_glass.xml, meshes with id's 5, 6, and 7 are excluded.

Homework 3

The images below were generated by using the supplied input files bunny.xml, horse.xml, killeroo.xml, dragon.xml, 

External libraries used:
- tinyxml2 (tinyxml2.h, tinyxml2.cpp are provided in the homework files.)
- lodepng (lodepng.h, lodepng.cpp are provided in the homework files.) 
- Eigen (Can be obtained from http://eigen.tuxfamily.org/) 

Command to compile:
g++ -I /usr/include/eigen3/ -std=c++11 -O3 main.cpp RayTracer.cpp Vec3d.cpp Ray.cpp shape.cpp tinyxml2.cpp lodepng.cpp BVH.cpp -o RayTracer

Computer specs:
Intel Core i7-2630QM CPU @ 2.00 GHz, 6 GB RAM

Running the ray tracer:
./RayTracer command should be fed by the relative path of the input file, e. g.,
795_input_set_02/horse.xml

This week's task was to accelerate render time by implementing Bounding Volume Hierarchy (BVH). 

File name: bunny.xml
Triangle count: 4967
Render time w/o BVH and flat shading mode:
~53 seconds
Render time w/o BVH and smooth shading mode: ~56 seconds
Render time w/ BVH and flat shading mode: 0.586 seconds
Render time w/ BVH and smooth shading mode: 0.604 seconds

File name: horse.xml
Triangle count: 11216
Render time w/o BVH and flat shading mode:
~32 minutes
Render time w/o BVH and smooth shading mode: ~40 minutes
Render time w/ BVH and flat shading mode: 7.643 seconds
Render time w/ BVH and smooth shading mode: 7.834 seconds

File name: killeroo.xml
Triangle count: 92092
Render time w/o BVH and flat shading mode:
Infeasable to wait
Render time w/o BVH and smooth shading mode: Infeasable to wait
Render time w/ BVH and flat shading mode: 1.822 seconds
Render time w/ BVH and smooth shading mode: 1.951 seconds

File name: dragon.xml
Triangle count: 871414
Render time w/o BVH and flat shading mode:
Infeasable to wait
Render time w/o BVH and smooth shading mode: Infeasable to wait
Render time w/ BVH and flat shading mode: 6.047 seconds
Render time w/ BVH and smooth shading mode: 10.928 seconds









Comments and Notes:

1. Interestingly, rendering horse.xml takes more time than killeroo.xml despite its triangle count. One possibility of this is the BVH tree may be constructed in an unbalanced way for horse.xml. The other possibility is there is a floor under the horse which makes box at the root of the BVH tree so big that almost every ray intersects with this box which makes intersection query times greater.

2. Although BVH accelerates rendering time tremendously there are few optimization techniques that can be applied:
 - Advanced ray-box intersection methods can be used instead of naive methods.
 - Advanced ray-triangle intersection methods can be used instead of naive methods.
 - Code level optimizations.
These acceleration methods will be implemented in the following weeks.

Homework 2

The images below were generated by using the supplied input files bunny.xml, horse.xml, simple_transform.xml, spheres_transform.xml and horse_instanced.xml.

External libraries used:
- tinyxml2 (tinyxml2.h, tinyxml2.cpp are provided in the homework files.)
- lodepng (lodepng.h, lodepng.cpp are provided in the homework files.) 
- Eigen (Can be obtained from http://eigen.tuxfamily.org/) 

Command to compile:
g++ -I /usr/include/eigen3/ -std=c++11 -O3 main.cpp RayTracer.cpp Vec3d.cpp Ray.cpp shape.cpp tinyxml2.cpp lodepng.cpp -o RayTracer

Running the ray tracer:
./RayTracer command should be fed by the relative path of the input file, e. g.,
795_input_set_02/horse.xml

The image on the left is the output of bunny.xml file with smooth shading mode. It illustrates how bunny looks when it is rendered with smooth mode.

Generated around 56 seconds.

Please note that same image with flat mode was generated around 53 seconds.

The image on the left is the output of horse.xml file with smooth shading mode and when no transformation is applied. It illustrates how horse looks when it is rendered with smooth mode and without any transformation.

Generated around 39 minutes and 54 seconds.



The image on the left is the output of simple_transform.xml file. It just includes objects in the file simple.xml from the previous homework with some transformations applied.

Generated below 1 second.











The image on the left is the output of horse.xml file with flat shading mode and with the given transformations applied. The transformations are applied directly to the object rather than the ray. We can infer from the figure that translation and rotation works correctly. However, there may be a bug with the scaling because the horse in the given reference output was not as tall as in this image.

Generated around 32 minutes.

 The image on the left is the output of spheres_transform.xml file. This time, instead of transforming the objects, the ray is transformed to the object space. The image does not match with the image provided as reference output. From sphere at the left in the image, we can infer that translation operation works correctly. However, most probably there is a bug with the scaling and rotation matrixes while transforming the ray into the object space.

Generated below 1 second.





The image on the right creates three instances of the horse in the image two above. While instancing, copies of the objects with desired transformations are created rather than transforming the ray. The horses in the reference output provided are smooth while in the image on the left are flat. This is due to the fact that I tested instancing before implementing smoothing and I the image on the left was generated. However, generation of this scene took so much time that I could not rendered the scene again with the smooth mode.

Generated around 118 minutes.