Hierarchical Reasoning for Vision Language Models (VLMs/LVLMs)

note-- api/ has reverse-engineered gpt-4V, DALL-E3 API's + SAM, SDXL, FUYU

Results/Example Slides--

// todo-- need to add steps to run here

Flow-- [Image -> Extract Segments (SAM) -> Extract Attributes + Caption] + Question ---> Answer

[Attributes considered (only relevant attributes extracted based on segment type out of these)-- (spatial relationships, pose estimation, depth estimation, motion, action, count, size, shape., color, coordinates, etc)]