After ChatGPT people, are inventing a lot of cool things and one of that amazing thing that I recently found is MiniGPT-4. In this post, I will explain you all about this new cool technology, MiniGPT-4, how it access and use it with real life examples.
What is MiniGPT-4?
MiniGPT-4 is a new model that can help you to understand any image. That means it can describe any image and can tell minor details of that image. It is a great alternative to ChatGPT 4. The model that MiniGPT-4 is using is Vicuna. Vicuna Is a great alternative to ChatGPT 4 and it has an accuracy of 90% of that of the ChatGPT 4.
Let me tell you the meaning of the MiniGPT-4 by giving a real life example. So suppose you went for dinner in a restaurant and there you saw an amazing image of a dish. Now you can just take that image and you can come home and just upload the image into the MiniGPT-4. Then you can just ask this model about the ingredients that you can use to create such a dish.
So basically MiniGPT-4 is just analyzing an image and it is giving us the correct answers of whatever questions we ask about that image. This makes MiniGPT-4 a really interesting AI model and that’s why it’s going viral on the internet.
Now let me show you the demo of this new AI model. We uploaded an image of a wolf and a lamp and we just asked this model to describe this image and you can see in the below image the generated output. It is describing the image correctly as well as it is giving us all the details like very micro level details of that image.
Example two of using this model. We uploaded this image of Elon Musk and we asked this model who is the person in the image. And it just gave the correct answer.
How to access MIniGPT-4?
To access MIniGPT-4, go to this website and try uploading an image. And then ask any questions.
How to install MiniGPT-4 on your local environment?
To start using MiniGPT-4 on your laptop or computer, just follow below steps:
Step 1: Setup code & environment
To clone the Github repository and to active use following lines of code:
git clone https://github.com/Vision-CAIR/MiniGPT-4.git
cd MiniGPT-4
conda env create -f environment.yml
conda activate minigpt4
Step 2: Setup Vicuna weights which are already trained
Click here to learn about Vicuna weights. Final weight structure should be as below:
vicuna_weights
├── config.json
├── generation_config.json
├── pytorch_model.bin.index.json
├── pytorch_model-00001-of-00003.bin
…
Step 3: Play with MiniGPT-4
Download the pretrained checkpoint form here. This will help you in using already pretrained model by creators of MiniGPT-4
Open config file line 11 and insert path of pretrained checkpoint
Is MiniGPT-4 better than the ChatGPT?
MiniGPT-4 and ChatGPT are totally different. MiniGPT-4 is a multimodal language model that generates outputs based on text as well as image. ChatGPT is a conversational language model that will generate text output for text input.
MiniGPT-4 is better for describing images or generating stories from images. Both models have their own pros and cons. But I think if you are doing a work in which you want like descriptions of the images then go for MiniGPT-4
Final Conclusion:
I think MiniGPT-4 is definitely worth to try. Still it is in development phase and I am pretty sure people are going to build some amazing apps based on this model.
Play with it and see how you can use it in your domain.