Question about custom map and GPT-4 Demo

[uk200200]

This work is truly inspiring.
I'm intrigued by your work and am interested in exploring your methods further.

Here are some questions, and I was hoping you could provide some guidance:

1. How can I use custom NeRF maps into your framework? Any specific requirements or steps would be appreciated.
2. Can you upload the code of GPT-4 demo for enhanced conversational capabilities?

Your insights on these matters would be very helpful for me in my research.

[XuweiyiChen]

1. Yes. So we preprocessed 3D data using LERF's code and saved it in the H5 format. The goal is to acquire a point cloud where each point contains a CLIP embedding at all scales. This step is just a tradeoff between efficiency and space.

Similarly, you can acquire a pair of data for OpenScene embeddings (which is a lot easier and better to use from my experience). The steps are described in the README.

2. The code for the GPT-4 demo is in the main branch. Let me know what does not work for you.