Machine Perception Project 2: Structure from Motion by John Han and Ashley Tsang

This is an implementation of Tomasi and Kanade's paper, "Shape and Motion from Image Streams under Orthography: a Factorization Method."

Setting up Dependencies and Running Code

We used Python 3.6 and conda environment for this project. To run our code and install the dependencies, please run

conda create --name env_name python=3.6
chmod +x dependencies.sh
./dependencies.sh

Afterwards, to run our project, go into the src directory and run

python main.py --data castle --points 800

to run Structure from Motion using the castle dataset or

python main.py --data medusa --points 800

to run SfM using the medusa video. In order to visualize the results, enter into the root directory and run visualize.py and specify visualization of dataset and motion or of shape. For example,

python src/visualize.py --dataset medusa --type R

would visualize the camera motion for the medusa dataset, whereas

python src/visualize.py --data castle --type S

would visualize the shape point cloud for the castle dataset.