Apr 18, 2023 20:00:00

Introducing 'Minigpt-4,' which lets you talk to AI about images

' Minigpt-4 ' has been released, combining '

Vicuna-13B ,' a chat AI said to have accuracy comparable to ChatGPT, and ' BLIP-2 ,' which can add captions to images, allowing users to have conversations with the AI about uploaded images.

Minigpt-4
https://minigpt-4.github.io/

On the Minigpt-4 announcement page, you can actually see the demonstration. First, click on the area in the red frame.

Next, choose the image you want to use as the topic of your conversation.

Click the 'Upload & Start Chat' button.

The upload will take a while.

When it's ready, the image and input field will reappear, so enter your question in the 'User' field.

The answer has been displayed.

You can reset the images and conversations by clicking 'Restart.'

At the bottom of the uploaded image, you can see the parameters 'beam search numbers' and 'Temperature.' 'Beam search numbers' are thought to be potential numbers for

beam search . Increasing this number tends to increase the number of answers, and even if you ask a question in Japanese, you will now get an answer in English.

'Temperature' refers to the diversity of responses, but even when I tried adjusting it, there was no noticeable difference.

At the time of writing, there was a high volume of access, so multiple mirror links for the demo were provided.

The source code is also available, so it can be run locally. Furthermore, a lightweight version that can run on a single GeForce RTX 3090 is also under development.

GitHub - Vision-CAIR/MiniGPT-4: MiniGPT-4: Enhancing Vision-language Understanding with Advanced Large Language Models

https://github.com/Vision-CAIR/MiniGPT-4