generative sketch to 3d application, where user can draw anything and have a 3d object of the highest quality generated. Bring your ideas to life
canvas3D-v1: This folder contains the first implementation of the project in which, user can sketch a drawing on a canvas and the program makes use of the OpenAI CLIP embedding model to identify the 3D object which best fits that sketch, text description and the selected style
canvas3D-v2: current iteration Makes use of the Openai point-e model to generate a 3D object for both the text and sketch inputted by the user. This model was choosen due to the fact that it can generate 3D objects in less than 3mins and with incredibly low compute time
https://github.com/openai/point-e
canvas3D-v2: In-development
Makes use of a pretrained generative model, to generate a 3D object that best fit the user sketch, text description, and selected style.
- Combine the text and image input using the openai CLIP mode
- Make use of the pretrainied Magic3DSketch model
- retrained the model on a new dataset "______" in other to increase range of the range of the model and its quality
- Improve quality during inferencing
install: node.js
npm
How to use: cd. v2-frontend command: npm run dev
pip install git+https://github.com/openai/point-e pip install fastapi uvicorn python-multipart pillow torch trimesh