search

Yushi-Hu / VisualSketchpad

Codes for Visual Sketchpad: Sketching as a Visual Chain of Thought for Multimodal Language Models
Apache License 2.0
106 stars 7 forks source link

May I ask where the matplotlib code in the input comes from? #2

Closed wwzhuang01 closed 2 months ago

wwzhuang01 commented 3 months ago

Great job! I have a question about the Geometry Problems in Section 4, Sketching to Solve Math Problems, where you mentioned that SKETCHPAD takes a geometry diagram and its corresponding matplotlib code as input. How did you obtain the matplotlib code for the images in the Geometry3K dataset?

Yushi-Hu commented 3 months ago

Thanks for your interest in our work! We will release codes and data in early July. Geometry 3K has the coordinates for each point, so we use GPT-4V + human edits to manually write all the codes for the images, which takes many human effort. We will update them soon!

va1shn9v commented 2 months ago

Hello, Amazing Project. I wanted to try this out, may I know when would the codes be released. Thanks for you work.

Yushi-Hu commented 2 months ago

we are working towards to release it in a week