ilsenatorov/kilter

Kilter DL

This repository has various half-finished models and scripts for dealing with KilterBoard climbs (generating/grading/autoencoding).

Installation

conda create -n kilter -f env.yaml

Preprocessing

Use the preprocess.ipynb notebook to preprocess the data, generating the necessary csv files in the data/ directory.

Diffusion model

Use the python diffuse.py command to train a diffusion model for climb generation.

Prediction ViT model

Use the python train.py command to train a model to predict the difficulty of the climb.

Sampling

Use the sample.ipynb notebook to load trained diffusion and prediction models together to generate climbs.