Home

Awesome

Home of Omega, a napari-aware autonomous LLM-based agent specialized in image processing and analysis.

License BSD-3 PyPI Python Version tests codecov Downloads Downloads napari hub DOI GitHub stars GitHub forks

<img src='https://github.com/royerlab/napari-chatgpt/assets/1870994/c85185d2-6d16-472d-a2c8-5680ea869bf2' height='300'> <img height="300" alt="image" src="https://github.com/royerlab/napari-chatgpt/assets/1870994/f3ea245e-dd86-4ff2-802e-48c2073cb6f9">

A napari plugin that leverages OpenAI's Large Language Model ChatGPT to implement Omega a napari-aware agent capable of performing image processing and analysis tasks in a conversational manner.

This repository started as a 'week-end project' by Loic A. Royer who leads a research group at the Chan Zuckerberg Biohub. It leverages OpenAI's ChatGPT API via the LangChain Python library, as well as napari, a fast, interactive, multi-dimensional image viewer for Python, another week-end project, initially started by Loic and Juan Nunez-Iglesias.

What is Omega?

Omega is an LLM-based and tool-armed autonomous agent that demonstrates the potential for Large Language Models (LLMs) to be applied to image processing, analysis and visualization. Can LLM-based agents write image processing code and napari widgets, correct its coding mistakes, performing follow-up analysis, and controlling the napari viewer? The answer appears to be yes.

The preprint can be downloaded here: 10.5281/zenodo.10828225

In this video, I ask Omega to segment an image using the SLIC algorithm. It makes a first attempt using the implementation in scikit-image but fails because of an inexistent 'multichannel' parameter. Realizing that, Omega tries again, and this time succeeds:

https://user-images.githubusercontent.com/1870994/235768559-ca8bfa84-21f5-47b6-b2bd-7fcc07cedd92.mp4

After loading a sample 3D image of cell nuclei in napari, I asked Omega to segment the nuclei using the Otsu method. My first request was vague, so it just segmented foreground versus background. I then ask to segment the foreground into distinct segments for each connected component. Omega does a rookie mistake by forgetting to 'import np'. No problem; it notices, tries again, and succeeds:

https://user-images.githubusercontent.com/1870994/235769990-a281a118-1369-47aa-834a-b491f706bd48.mp4

In this video, one of my favorites, I ask Omega to make a 'Max color projection widget.' It is not a trivial task, but it manages!

https://github.com/royerlab/napari-chatgpt/assets/1870994/bb9b35a4-d0aa-4f82-9e7c-696ef5859a2f

As LLMs improve, Omega will become even more adept at handling complex image processing and analysis tasks. GPT 4.0 has been a significant upgrade compared to GPT 3.5, and many of the videos (see below and here) are highly reproducible, with a typically 90% success rate (see preprint for a reproducibility analysis). While open-source models are promising and rapidly improving, they must get better to run Omega reliably. More recent models by OpenAI's competitors, such as Google and Anthropic, are great news, but Omega still needs to support these newer models fully -- it seems every week comes with a new batch of models.

Omega could eventually help non-experts process and analyze images, especially in the bioimage domain. It is also potentially valuable for educative purposes as it could assist in teaching image processing and analysis, making it more accessible. Although ChatGPT, which powers Omega, may still need to be on par with an expert image analyst or computer vision expert, it is just a matter of time...

Omega holds a conversation with the user and uses different tools to answer questions, download and operate on images, write widgets for napari, and more.

Omega's Built-in AI-Augmented Code Editor

The Omega AI-Augmented Code Editor is a new feature within Omega, designed to enhance the Omega's user experience. This editor is not just a text editor; it's a powerful interface that interacts with the Omega dialogue agent to generate, optimize, and manage code for advanced image analysis tasks.

<img src='https://github.com/royerlab/napari-chatgpt/assets/1870994/cf9b1c15-f11a-4e25-a73d-d96915c46c6a' width='800'>

Key Features

Usage Scenarios

You can find more information in the corresponding wiki page.


Omega's Installation instructions:

Assuming you have a Python environment with a working napari installation, you can simply:

pip install napari-chatgpt

Or install the plugin from napari's plugin installer.

For detailed instructions and variations, check this page of our wiki.

Requirements:

You need an OpenAI key; there is no way around this, I have been experimenting with other models, including open-source models, but right now, the best results, by far, are obtained with ChatGPT 4 (and to a lesser extent 3.5). Check here for details on how to get your OpenAI key. In particular, check this for how to gain access to GPT-4 models.

Usage:

Check this page of our wiki for details on how to start Omega.

Tips, Tricks, and Example prompts:

Check our guide on how to prompt Omega and some examples here.

Video Demos:

You can check the original release videos here. You can also find the latest preprint videos on Vimeo.

How does Omega work?

Check our preprint here: 10.5281/zenodo.8240289 and our wiki page on Omega's design and architecture.

Cost:

Developing the initial version of Omega cost me $13.97, hardly a fortune. OpenAI pricing on ChatGPT 4 is very reasonable at 0.01 dollars per 1K tokens, which means $1 per 750000 words.

Note: you can limit the burn rate to a certain amount of dollars per month, just in case you let Omega think over the weekend and forget to stop it (don't worry, this is actually not possible).

Disclaimer:

Do not use this software lightly; it will download libraries of its own volition and write any code it deems necessary; it might do what you ask, even if it is a very bad idea. Also, beware that it might misunderstand what you ask and then do something bad in ways that elude you. For example, it is unwise to use Omega to delete 'some' files from your system; it might end up deleting more than that if you are unclear in your request.
Omega is generally safe as long as you do not make dangerous requests. To be 100% safe, and if your experiments with Omega could be potentially problematic, I recommend using this software from within a sandboxed virtual machine. API keys are only as safe as the overall machine is, see the section below on API key hygiene.

THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

API key hygiene:

Best Practices for Managing Your API Keys:

Contributing

Contributions are extremely welcome. Tests can be run with tox, please ensure the coverage at least stays the same before you submit a pull request.

License

Distributed under the terms of the BSD-3 license, "napari-chatgpt" is free and open-source software

Issues

If you encounter any problems, please file an issue along with a detailed description.