Virtual Showroom

Note: I am not very good at English. This README.md was proofread by ChatGPT.

Background

Showrooms can be costly, but their ROI is unclear. Therefore, companies generally aim to achieve semi-automation using Multimodal AI.

Virtual Showroom Concept

The Virtual Showroom is highly cost-effective. It only requires a 240-degree panoramic screen to provide VR experiences without the need for headsets. I aspire to create a "real" virtual showroom at work someday, but for now, this project is focused on developing a "virtual" virtual showroom using Unity, either as an AR app or a console app, as part of my hobby project.

=> Demo video on YouTube

Reference Showroom

I visited NISSAN GALLERY at Nissan global headquarters in Yokohama in March 2024. This was for a driving experience in harsh environments.

Schematic

My original design for a virtual showroom.

The 240-degree panoramic screen in this project also supports perspective drawing, similar to the driving experience at the NISSAN GALLERY mentioned above.

System Architecture

The Flask-based API server runs on Raspberry Pi, suitable for home LAN environments.

Development Environment

Tools

• Blender with MPFB2 (Blender)
• Unity
• VS Code

LLM

This project utilizes OpenAI's "gpt-4o-mini" with Multimodal RAG (text and image).

Scenes

This project currently has two scenes:

• Virtual Showroom (it also runs as an AR app on a mobile phone)
• Object Detection

More scenes will be added in the future.

Virtual Showroom scene

1. Designing Models (Blender)

=> MODELS.md

2. Creating the Virtual Showroom (Unity)

=> SHOWROOM.md

3. Developing the API server

The API server runs on Raspberry Pi.

Run the "app.py" that is in my another repo compact-rag.

4. AI-Controlled Virtual Showroom

=> AI_Controlled.md

Object Detection / image recognition scene

Although this scence has nothing to do with RAG, I use the Flask-based API server for simplicity.

AI Robot

I was impressed by the demo video for Figure 01. If such a robot were to exist, human guides in showrooms might no longer be necessary.

I modified the robot included in Unity's Starter Assets to experiment with what Generative AI can achieve.

Experiments

A few years ago, I developed several image recognition and object detection apps using TensorFlow like this one.

This time, I’m utilizing OpenAI's multimodal AI for image recognition and object detection.

I uploaded the experiment result to YouTube: The experiment of multimodal AI with Unity

After trying various experiments, I found that an image size of at least 512px x 512px seems to work best.