I dont see how this project breaks from that though. It looks pretty useful for quickly visualizing information common in computer vision. What "does not see your way"?
You're absolutely right that Streamlit isn't a Jupyter replacement (we're not trying to be!), in the same way that Spyder isn't a "replacement" for the JetBrains set of IDEs. Maybe Streamlit will one day move to multiple language support, but for now, we're trying to be the best Python library we can be for creating web apps.
As far as GPU-support goes, as mentioned above, Streamlit "supports" GPU in as much as Python does. It's important to separate out front-end/UI from backend. If something is supported with GPU as a backend, "data science" (however defined) functionality, Streamlit supports that in the same way a React app calling a GPU API endpoint or other workflow is "GPU-enabled"
GPU support is not. You can use GPUs with Streamlit, since its app scripts are normal Python code just executed through Streamlit framework.
Jupyter and Streamlit are not competitors, but they covers the needs the other lacks.