Hello!
I am a computer vision and agricultural robotics researcher, working on methods for optically scanning, reconstructing and extracting structure of plants.
Interested generally in functional programming, machine learning and various forms of 3D reconstruction, SFM, photogrammetry etc.
Also keen on climbing, running and cycling - proportions of which vary depending on enthusiasm levels!
Christchurch, New Zealand
A Photorealistic Painting of Christopher Luxon is now A Redacted Painting of Christopher Luxon. Let's see if it stays up this time.
As before, please ask me Questions on the auction listing, so I may Answer them
https://www.trademe.co.nz/a/marketplace/art/nz-artists/listing/4974376165
@aligorith (Sometimes) I wish autocorrect would learn when you went back and fixed one of it's "corrections" that you clearly meant it, rather than suggesting the same thing over again!
The two most time-saving uses of #AI I am finding in my work are not in the obvious use case of asking AI about my math problems, but rather either in creating computer code for various scientific computations, or in mundane data formatting tasks. A case in point is that of putting together a bibliography for a paper (with or without BibTeX). Often, I need to grab a dozen or so references from other sources (e.g., PDF documents) and format them to a specific bibliography style. This is not difficult, but is certainly tedious. With Github Copilot, it has become quite trivial; I paste in the raw reference in some random format as a comment, and Copilot usually understands without any additional prompting that I want to convert it to precisely the same format that all other references are in, usually correcting for any typos and other misprints in the original comment This is far faster than any previous method of collecting references that I have used. (And I have started taking joint papers in Overleaf offline to edit, solely to take advantage of these sort of autocompletions.)
@kurtsh Seems it's mostly just a writeup of the pet peeves of one academic.
If text is so inadequate as a method for understanding then why does the author bother writing an article? It's just text after all.
@aligorith USB (camera) connectors and cables are the bane of my life - still way better than ethernet, though cat5 connectors would be nice!
@aligorith Depends on the port and cable, some of them fit very snugly - there's definitely some variation in how they're made...
Test of some alternative split and prune operations on 2D image. Black and white image to give a sparse scene (more similar to 3D).
A little #introduction if I may:
I'm Buck, by day I work on self driving cars and by night I work on #OpenSource. I'm currently focused on @formak to magically translate Python to high performance C++. You can read more about FormaK here https://buckbaskin.com/blog/tag/formak.html
(edited to link the official FormaK Mastodon)
@amytabb That does sound like something typically painful! Do you have to seek out funding for the work you want to do? Does it revolve around some other outcomes?
We typically need to seek government grants for our work, usually it has to revolve around some supposed commercial outcomes. We were lucky enough this year to get a new one to continue what we were doing anyway - but it could have looked quite different.
@KiffinEileen Real problems!
@brohrer My home is just cluttered with things everywhere and not enough space to put them. Any empty space anywhere is very welcome!
It's also good for what I originally wanted! Which was de-coupling projection from rasterization and allowing any sized features without recompile. Using Taichi for implementation gave me both, maintainable projection using autograd and easy JIT for parameter changes.
Still much to do, applying to a 3D training scenario rather than just dummy 2D tests (like the optimizing-to-image in the repo), and of course a variety of experiments for which I wanted it in the first place.
I put a small write up I put a small write up here:
https://github.com/uc-vision/taichi-splatting/blob/main/BENCHMARK.md
The benchmark just renders original camera placements (forward and backward).
In particular this implementation seems to be much faster at the backward gradient pass (still the most expensive part, but less).:
Over the last few weeks I took the taichi_3d_gaussian_splatting code and refined it to try to make it more modular, for some experiments I want to try. I found some optimisations. Here's some benchmarks vs. the original, and official GS implementation. IMO it turned out nicely.
#GaussianSplatting #DifferentiableRendering #ComputerVision #Rendering
2024: "We have respirators that electrostatically capture viral particles, ventilation and filtration systems to clean indoor air, and UV lights to kill viruses."
1918: "That's incredible! What's it like living without all this illness?"
2024: "Oh, we don't actually use them."
How the death of Twitter and the continuing enshittification of X is reviving the academic blogosphere:
https://jabberwocky.weecology.org/2024/01/08/return-to-and-of-the-ecology-blogosphere/
and
https://dynamicecology.wordpress.com/2024/01/08/dynamic-ecology-is-back/
The planet-warming emissions generated during the first two months of the war in Gaza were greater than the annual carbon footprint of more than 20 of the world’s most climate-vulnerable nations, new research reveals.
Another reason why Greta is, of course, right. No #ClimateJustice without peace
