Suppose you were a funder wanting to design a system to fund science projects that were bottom up rather than top down. How would you do it?
I think you'd want to restrict it to non-faculty to start with, and have some sort of consensus-building rather than competitive approach. Like, maybe you could have an initial round where people proposed ideas, followed by a second round where people indicated who they'd be willing to work with and which aspects of their ideas they'd be willing to drop or modify in order to build consensus. Possibly you might need multiple rounds like this until you iterated on a solution that worked.
Would there by problematic hidden power dynamics in an approach like that? I guess so, there always are. But maybe still better than top down approach?
And is there any chance of finding a funder who would be willing to experiment with such an idea? Or any existing examples of experiments like that? Or more generally, examples of funders taking a non-competitive approach?
To non-faculty for sure. My first move would be to expand funding for PhD students: attract many, and with a good salary to bias the choice away from industry.
Despite the fact that IMOD insists it can read 16-bit MRC files, I have found that header...can't. I wrote quick and dirty python version to get the information I'm usually looking for (I'm sure smarter people will have opinions, but this works).
"I would have never thought moving away from a 10yr old JDK could be this smooth!" – Tiago Ferreira, author of the SNT plugin for neuronal tracing among others.
Curtis Rueden pushing forward the release of #FijiSc with #Java21 – a huge upgrade from the decade-old java 8 that Fiji uses today.
Testers are reporting success even in new MacOS chipsets.
I tried opening an RStudio project from 2017 for which I used {packrat}, and I was disappointed—but not surprised—that it didn't work. I'm becoming convinced that "reproducible analyses" are a pipe dream. #RStats
I routinely run #java code we wrote in 2005–2012. And scripts in jython on top of that written from 2010 onwards. All in #FijiSc – https://fiji.sc for image processing.
Perhaps the #RStats community does not value long-term stability or hasn't adopted backwards compatibility strategies when updating libraries?
The #neuroscience Mastodon hive-mind worked wonderfully earlier - let’s try again!
In #napari, I have loaded a video using the napari-video plugin. I would like to get some simple image statistics for all frames in the video - does anyone know how to? I have looked through the various plug-ins, but none seem obvious… but it’s so simple (no segmentation, nothing complicated), I feel like I must be missing something!
If the insect-on-a-ball video loads into #FijiSc with the ffmpeg plugin, then I'd use the optic flow plugin, extract vectors for all frames, and detect when vectors get larger than a defined threshold to avoid noise.
Go to "Help - Update...", then in the menu that opens click on the button labeled "Manage Update Sites". Then search for "FFMPEG" and tick its checkbox on the left, and click on "Apply and close", then apply the updates, and when done then restart #FijiSc. The plugin will appear under "File - Import - Movie (FFMPEG").
Then with the movie open, go to "Plugins - Optic Flow" and choose one of the options ("Integral Block MSE", "Integral Block PMCC", "Gaussian Window MSE").
Biologists! I received a large .tiff image from electronic microscope imaging. It doesn't show well in any image viewer I tried on Mac. The thumbnail is perfect though, so I know the image is fine. What should I do to see it, or to convert .tiff to .png properly (the usual conversion methods haven't worked)
A good friend of mine had an MRI, but the insurance wouldn't pay for a doctor to process/analyze it. First of all I didn't know you could even do that, and second of all do I reach any neurologists or otherwise the medical kind of brain doctor that would be willing to take a pro-bono look at an MRI?
edit: we have figured something out! thank you all so much for your offers and for spreading the word, it means a lot that people are willing to help out a friend in need like this, consider my heart warmed <3
Why would one want to run machine learning inference from #java?
To do so on 3D, 4D, ND datasets, trivially accessible from image processing and visualization libraries such as #ImgLib2, the #BigDataViewer, #LabKit and more, all integral parts of #FijiSc.
i was playing an in-browser game with someone last night and got to talking and i mentioned i was a programmer. they said they had always wanted to learn, and so i suggested the time tested strategy of just getting mad whenever a computer didn't do something they wanted it to. they mentioned hating some features of the game, and since the game's code is clear (semi-retro Angular) and comes with a sourcemap i recommended they try taking a stab at reading it. i gave them some tips on starting points and how to read it. we were playing a game so i wasn't necessarily trying to teach, just give them a starting point for later if they wanted.
they told me that they would try but the learning curve was always a cliff, and i knew they were right but holy SHIT they were so immediately right. they asked me how to save the source to read later, and there isn't actually a way to save unpacked sourcemap code from firefox. I said your best bet was to just copy and paste each file, or since they already had homebrew and npm, they could try copy/pasting something like wget URL; npx unpack file; open . and linked them to this package which to me is super clear, but they were just dumbfounded by it.
They were (correctly) like "you say follow the instructions there but the word "instructions" is nowhere on this page, literally the entire thing is code. there isn't even a download button or anything, how do i get the program, what the fuck is this site?" Every part of what i was telling them was totally new to them, I thought that "just open this program and copy and paste this text" would be doable even if they didn't get what was going on, just so they would have something to look at later, but that act of exposure was so discouraging I felt bad and we just quit the game and i talked them through some of it. We got stumped at merely trying to download the code. Not even reading it yet, definitely not trying to run it.
I've taught ppl to program to a level of basic self-directedness maybe a dozen times, and every time I remember just how inaccessible this whole racket is. I remember all that extremely well myself, and I still am not close to being able to imagine what a really legible programming ecosystem would look like
I find writing documentation relaxing. It's also the best way I have to future-proof my own work: so that I know how I did what down the line. For example, see this labour of love over 13 years, for image processing in #FijiSc: https://syn.mrc-lmb.cam.ac.uk/acardona/fiji-tutorial/ As far as I know all the scripts run to this day, and it's proven invaluable time and again to myself – and likely to others, which is a win-win.
Scientists will be like "results should be replicable!" but then do all their experiments with a random walk of homebrew code that runs on four computers networked with a nest of BNC cables, each with a different version of MATLAB, and after every experiment the data is saved by walking a flash drive around to each of them since they cant be connected to the internet because one of them still runs Windows XP and if the rest so much as heard of a software update the work of 5 grad students whose whole PhD was spent setting up this monstrosity would be ruined forever.
"A problem often related as 'the computer science PhD student moved on, and we do not know what parameters were used, neither what the magic numbers mean'."
"Compared with the honeybee and the fruit fly, Megaphragma exhibits the following miniaturization-related adaptations: a significant reduction in the number of ommatidia, absence of several cell types, reduced size, and denucleation of neurons. Interestingly, the reduction in lens diameter is less than that expected from the optimization of the optical resolution of the eye. This suggests that light sensitivity is a more important
consideration when lens diameter approaches the wavelength of light. The absence of wide-field (or non-columnar) lamina neurons in Megaphragma could be a consequence of the smaller number of ommatidia, their larger acceptance angle, and the lower resolving power of the eye."
Volume assembled with #FijiSc and #TrakEM2, and its neurons and synapses mapped with #CATMAID. Woohoo!
For any #3DEM#volumeEM people: is there an alternative software for skinning that you can recommend?
I’m building a 3D model of a complex subcellular structure, using 3DMOD. During skinning, imodmesh is generating a many misconnections. I've minimised them but I’m wondering if other software has better algorithms for generating the mesh.
May I suggest #FijiSchttps://fiji.sc which is a free and open source image processing software. Handling very large images should not be an issue. We handle volumes of many terabytes with it. Can open and write pretty much any file format imaginable, including raw, and offers a huge amount of image filters.
bioRxiv submissions seem to have hit a plateau of 3K/month. Wonder if this represents competition from other (commercial) preprint servers, post-pandemic effects or something else.
@quantixed Coincidentally, this year the #FijiSc software paper’s citations plateaued, after 10 years of growth (published in 2012 https://www.nature.com/articles/nmeth.2019 ), as if there weren’t any more labs that could start to cite it. Knowing the size of biology academia would be interesting.
Can an #LLM solve esoteric #programming problems, like camera motion in #Blender3D? So far, I'm seeing only limited success. The GPT-4 (advanced) model of phind.com can solve: "Write Python using the Blender API to animate orbiting of the camera by 90 degrees around the Z axis centered at object 'A' from frame 1 to 24." It parents to a pivoting empty node, a good approach. But the GPT-3.5 (standard) model gets confused. And the GPT-4 model fails on a more useful version of this problem. (1/3)