CMPT 464 Final Project - Chair Generation

Github link: https://github.com/pmiiller/chairGeneration

Instructions

• Run python clean_meshes.py in order to clean the training data
  • Must be run before anything else!
• python main.py to generate a single chair with a random template
• python main.py <chairDirectoryName> to generate a single chair with a specified template
• python main.py all to generate a chair for all templates and then score them
• python main.py eval to evaluate a sample of already generated chairs
• python main.py load to load all templates again and create a new pickle file
• python main.py cluster to create clustering of the parts and create a new pickle file of parts
• python main.py scorer to run the validation for the scorer
  • This runs the evaluation script on chairs generated using our method which we separated into two sets; "good" chairs and "bad" chairs. Good chairs on average should have a higher score than bad chairs.
• Important: After running scorer or modifying the input chair data, you need to empty the new_chair_bmp, new_chair_obj, and ranked_chair_obj folders before running all or the jupyter notebook. Not doing so may lead to unexpected behaviour.

Download the LeChairs scorer from here.

Requirements:

• Python 3.*
• LeChairs
  • Modified versions of the LeChairs training and evaluation scripts are already included in the repo
• Trimesh
  • Installation:
    • pip3 install trimesh
    • show() requires pyglet: pip3 install pyglet
    • trimesh.registration package requires rtree pip3 install rtree, sudo apt install python3-rtree
• Run pip3 install -r requirements to install the needed pip requirements