GPT-4V may be surprisingly robust here. Set of mark prompting(which is accomplished here with Vim) improves grounding by a silly high amount. https://som-gpt4v

GPT-4V may be surprisingly robust here. Set of mark prompting(which is accomplished here with Vim) improves grounding by a silly high amount. https://som-gpt4v.github.io/ https://som-gpt4v.github.io/