Climb GPT is an application that uses Machine Learning to do perform grading of climbs. Currently looking at how to use segment-anything library by FAIR to segment routes and then perform classification.
- How to do prompting
- Format of the masks that is returned by the model
- Running of Model iwth ONNX run time
- Incorporate model with frontend
- Simple problem of segmenting a climb into different components based on the colour of the holds
Things to do:
- Grade Tags
- All handholds and footholds
- Clustering same coloured routes together
- Hold Classification
- Wall Angles
- Grading of climbs
- Beta generation
- Step by Step format
- Isolate the different routes by colour
- Identify the different types of holds
- Identify the type of terrain
- Accumulate the information about the different holds
- Perform generation using GPT model
- Able to submit route posts
- What is the schema for the database
- Cloud Architecture for the database
- Provide information about the routes
- Perform clustering
- Cluster routes which are the same together
- Allow people to submit their own videos
- Generate video on how to send the route
- Translate the GPT output into a video guide using DALLE/ Diffusion Models
- How to use ONNX with Browser runtime
- Preprocessing Pipeline
- Patch Size
- Prompting - How to edit prompting for the model