YouTube-like service but for microscopy (+ bonus)

Hi all,

I’d like to share an early preview of the project I’m working on: an YouTube-like service but for microscopy called Membrane. The focus is on seamless sharing of the data online without the fuss of permanent storage (like in IDR). It’s meant to aid people working day-to-day with microscopy data. It came out of a collaboration with a bio lab (my background is in software) and their need to work online in the face of COVID19 restrictions. I know many labs struggle with moving to work online so I thought Membrane is worth sharing more widely.

I recorded a demo: Membrane demo - YouTube

The online viewer came out of necessity of the moment. The main problem I’m tackling with Membrane is a “hands-free”, generalist segmentation service, where you upload a microscopy image of any kind and get a segmentation back. I know this is a hard problem but Nucleaizer and Cellpose made great strides in this direction and we’re heavily inspired. Our focus is on people who do not want to fiddle with GPUs or figure out strategies of scaling to large datasets. The video contains a sneak preview of that service. A few biologists who tried it were deeply impressed.

I’d love to hear what you think! We’re not quite ready to open it up to the internet, but if you would like to try it let me know here or at


In an attempt to be brief, I forgot to give credits: the renderings you see in the demo are powered by napari - we’re big fans.


Can you give some technical details? I’ve been thinking of trying something similar in the context of the EOSC-Life project. Along a similar line, we’ve started dealing with remote image analysis work by setting up a virtual desktop system. It’s still in beta but you can try it.

Hi Jean,

Depending on what you’re asking. Membrane runs in the browser and heavy computations (e.g. segmentation) are all offloaded to cloud (AWS), including running napari in headless mode. From what I have seen, biologists tend to have “bursty” workflows with large amounts of data generated upon finishing an experiment spaced by quiet periods. Ideally, you would like to process the data instantly. This a classic example where cloud computing shines and I’ve developed a belief that biologist would immensely benefit from fully embracing cloud tools.

Curious what made you go the virtual desktop route. Is it because all existing pieces of software are desktop ones?

Curious what made you go the virtual desktop route. Is it because all existing pieces of software are desktop ones?

Yes, at least in part. Another reason is that in some cases we need an interactive environment with access to different tools/software in particular for exploratory work.
I think this is a different use case than running computation-heavy stuff in an HPC/cloud. Note that it is also possible to send jobs to the HPC back-end with the virtual desktop.

Membrane runs in the browser and heavy computations (e.g. segmentation) are all offloaded to cloud

Have you considered using the Galaxy platform? It’s basically a web front end to run compute jobs elsewhere so sounds similar to what you do but with the added benefit of documenting the workflows.

For the exploratory work, people in the industry (e.g. tech companies) heavily rely on Jupyter notebooks or Spark notebooks (for big data-style computing). I found that setup to be really good. In Spark case, sending workload to a cluster of machines is the default and transparent to the user.
I wonder if it would make sense in this case? Or are we back to the square one: these different tools are GUIs and not libraries so you can’t use them in a notebook?

Re: Galaxy, no, I haven’t heard of it. The initial impression I get is that it might be a tad too complex to use for the lab I’m collaborating with.

On the one hand we’re talking about users who’ve never heard of Jupyter and on the other users who want to use the tools they are used to e.g. CellProfiler, ImageJ… all GUI-based.

Galaxy can look intimidating at first mostly because there are so many tools to choose from but it seems people get quickly comfortable with it.

Good point on the confusing of the two groups! My implicit bias is that I envision biologists adopting Jupyter-style workflow. I think of lightweight programming as a super power and jupyter delivers just the right format provided all tedious issues of running notebooks, installing dependencies, etc. are taken care of. This is definitely in the dream category but tools like point in that direction.

I envision biologists adopting Jupyter-style workflow

Twenty years ago I thought all biologists would realize the place that computers had taken in the field and would soon all be programming their way through their data. We’re far from there yet so I won’t hold my breath.
I’ve seen deepnote before but at the moment it doesn’t run where the data is. You don’t want to move multi TB size data sets around in the cloud and I find it ridiculous to have to do i/o over http.