16-889 Assignment 1: Rendering Basics with PyTorch3D

0. Usage of main.py

The main file can be run for each question to generate the results. The following is the description for each argument:

--question : The question number can be given from the list: ["1.1", "1.2", "2.1", "2.2", "3", "4.1", "4.2", "4.3", "4.4", "5.1", "5.2", "5.3", "6", "7.1", "7.2", "7.3", "7.4"]

--image_size : The image size can be specified. For example: 256, 512.

--num_samples : In case of point clouds rendering, we can specify the number of samples.

--output_path : The output gif or images are stored in the path given through this argument.

Example: python main.py --question 1.1 --image_size 256 --output_path outputs/dolly.gif

1. Practicing with Cameras

1.1. 360-degree Renders (5 points)

To render the 360 degree view for the cow, run the following command:

python main.py --question 1.1 --image_size 256 --output_path outputs/cow_360.gif

The output gif can be found in outputs/cow_360.gif

1.2 Re-creating the Dolly Zoom (10 points)

To render the Dolly zoom effect for the cow, run the following command. The output gif can be found in outputs/cow_360.gif

python main.py --question 1.2 --image_size 256 --output_path outputs/cow_360.gif

2. Practicing with Meshes

2.1 Constructing a Tetrahedron (5 points)

To render a tetrahedron, run the following command:

python main.py --question 2.1 --output_path outputs/tetrahedron.gif

The mesh has four vertices and four faces. The output gif can be found in outputs/tetrahedron.gif

2.2 Constructing a Cube (5 points)

To render a cube, run the following command:

python main.py --question 2.2 --output_path outputs/cube.gif

The mesh has eight vertices and twelve triangle faces. The output gif can be found in outputs/cube.gif

3. Re-texturing a mesh (10 points)

To render a custom textured cow mesh, run the following command:

python main.py --question 3 --image_size 256 --output_path outputs/retexture.gif

The colors chosen for the below texture are color1 = [0, 0, 1] and color2 = [1, 0, 0]. The output gif can be found in outputs/retexture.gif

4. Camera Transformations (20 points)

To render different camera transformations, run the following commands corresponding to the question numbers. The outputs can be found in outputs/camera_transform1.jpg, outputs/camera_transform2.jpg, outputs/camera_transform3.jpg and outputs/camera_transform4.jpg.

python main.py --question 4.1 --image_size 512 --output_path outputs/camera_transform1.jpg

python main.py --question 4.2 --image_size 512 --output_path outputs/camera_transform2.jpg

python main.py --question 4.3 --image_size 512 --output_path outputs/camera_transform3.jpg

python main.py --question 4.4 --image_size 512 --output_path outputs/camera_transform4.jpg

Camera Transformation	Image output
Rotate 90 degrees left along Z-axis (anti-clockwise)
Translation along negative Z-axis by 5 units
Translation along negative X-axis and positive Y-axis by 0.5 units
Rotation along Y-axis by 90 degrees, translation along negative X-axis and Positive Z-axis by 3 units

5. Rendering Generic 3D Representations

5.1 Rendering Point Clouds from RGB-D Images (10 points)

To render the pointclouds of the plant image, run the following command. The outputs can be found in outputs/pointcloud1.gif, outputs/pointcloud2.gif and outputs/pointcloud_combined.gif

python main.py --question 5.1 --output_path outputs/point_cloud.gif

5.2 Parametric Functions (10 points)

To render torus pointcloud with 100 samples, run the following command. The output gif can be found in outputs/torus.gif

python main.py --question 5.2 --num_samples 100 --output_path outputs/torus.gif

To render torus pointcloud with 1000 samples, run the following command. The output gif can be found in outputs/torus_1000.gif

python main.py --question 5.2 --num_samples 1000 --output_path outputs/torus_1000.gif

5.3 Implicit Surfaces (15 points)

To render an implicit torus, run the following command. The output can be found in outputs/torus_implicit.gif

python main.py --question 5.3 --output_path outputs/torus_implicit.gif

Tradeoffs of Mesh Vs Point cloud:

Rendering speed: It is faster to render a mesh compare to a point cloud.

Rendering quality: Mesh has higher resolution comapre to a point cloud( depends on the number if samples).

Ease of use: Point clouds are easier to use than mesh.

Memory usage: In case of meshes, we store the faces and vertices but in case of point clouds we need to store position of each point. Meshes are memory efficient in comparison to point clouds.

6. Do Something Fun (10 points)

For this assignment, I have rendered a Mario game simulation. The Mario character jumps over the obstacles in the game. I have placed the obstacles (cubes) and textured with shades of green. The Mario character has been textured with a gradient of red to blue. In this gif, I am translating the camera along the negative x-axis and translating the Mario character along negative x-axis and positive/negative y-axis to simulate jumping action.

To render mario, run the following command:

python main.py --question 6 --image_size 256 --output_path outputs/mario.gif

(Extra Credit) 7. Sampling Points on Meshes (10 points)

To render the pointcloud from cow mesh, run the following commad for 10 samples, 100 samples, 1000 samples and 10000 samples respectively. The outputs can be found in outputs/cow_pointcloud10.gif, outputs/cow_pointcloud100.gif, outputs/cow_pointcloud1000.gif and outputs/cow_pointcloud10000.gif

python main.py --question 7.1 --image_size 256 --num_samples 10 --output_path outputs/cow_pointcloud10.gif

python main.py --question 7.2 --image_size 256 --num_samples 100 --output_path outputs/cow_pointcloud100.gif

python main.py --question 7.3 --image_size 256 --num_samples 1000 --output_path outputs/cow_pointcloud1000.gif

python main.py --question 7.4 --image_size 256 --num_samples 10000 --output_path outputs/cow_pointcloud10000.gif

Cow Mesh	Point clouds 10	Point clouds 100	Point clouds 1000	Point clouds 10000