This repository contains the VG150 dataset transformed into datasets format with keys: "image_id", "image", "prompt_open", "prompt_close", "objects", and "relationships".
It was presented in the paper Compile Scene Graphs with Reinforcement Learning.
Code:
https://github.com/gpt4vision/R1-SGG
