fouronnes3

Torch Lens Maker - https://victorpoughon.github.io/torchlensmaker/

Open-source differentiable geometric optics in PyTorch.

Spectromers don't measure the subjective perception of color.

How would you do traits in Python?

Changes channel faster than a modern tv

Truck drivers going home and playing truck simulator.

Little reminder that there once was a "C++ error message competition" [0] where your score was measured as "the size of the error message divided by the size of the source code". The winning entry had a score of 5.9 billion.

[0] https://tgceec.tumblr.com/post/74534916370/results-of-the-gr...

Capsules do generate lift by having a slightly offset center of mass and control its orientation by rolling the entire capsule. This is used since the Apollo days to steer to the landing site. Even mars rovers use capsule lift during mars entry to steer.

Or move the moon closer.

Thanks. Yeah that would be great I'm sure. The issue I have is also getting everything started. I've got a bit of a chicken and egg problem where I need users to fund the development, but I need to develop more features to get users.

I'm really interested in this discussion. I started an open-source project [1], and I've been working on it for the past 6 months full-time on my own time and money (savings + benefits). First I was part of a batch at the Recurse Center [2], now I'm alone at home. I'd love to keep working on it because I have a big roadmap and ideas, but I'm not sure if I will be able to. It's heartbreaking to have to stop at version 0.0.1 when I know that 1.0 could be so good. If I just had the time...

[1] https://victorpoughon.github.io/torchlensmaker/

[2] https://www.recurse.com/

    Location: France
    Remote: Yes
    Willing to relocate: No
    Technologies: Python, PyTorch, C++, Linux, Webdev, Typescript, I like it all really - love learning new stuff
    Résumé/CV: https://www.linkedin.com/in/victorpoughon/
    Email: first name <dot> last name <gmail dot com>

Hire me to work on Torch Lens Maker [1], an open-source Python library for differential geometric optics. The goal of this project is to design optical systems with modern numerical optimization (PyTorch autograd). The website has a detailed roadmap of what I have planned.

I'm also happy to work on other stuff, I had a recent blog post [2] do fairly well on HN a few days back, which would give you get a great idea of how I work.

[1] https://victorpoughon.github.io/torchlensmaker/

[2] https://victorpoughon.fr/i-tried-making-artificial-sunlight-...

I'm working on Torch Lens Maker [1], an open-source Python library for differential geometric optics. The goal of this project is to design optical systems with modern numerical optimization (PyTorch autograd). The website has a detailed roadmap of what I have planned.

I also had a recent blog post [2] do fairly well on HN, and now I'm kinda thinking about a new design for version 2 of that project because I can't help it I guess.

[1] https://victorpoughon.github.io/torchlensmaker/

[2] https://victorpoughon.fr/i-tried-making-artificial-sunlight-...

I wasn't aware this existed, thanks! Very interesting.

Very cool! I definitely want a spectrometer at home now! edit: looks like it's more of a DIY project than a commercial thing. Maybe DIY spectrometer is my next project then!

I agree, thanks for underlying that point. I did not intend to write an ad for JLCPCB, but honestly I kinda did. I feel like it's deserved praise because the service is really great. Low volume custom PCBs with assembly service and 5 axis CNC with a minimum order quantity of 1 is game changer for prototyping. Combined with a software first approach to design with tools like build123d, the flexibility of 3D printing and 5 axis CNC, and the addictiveness of the high dopamine feedback loop of software dev starts to bleed over to hardware dev. Shipping delay is the last hurdle.

About the lenses, I did consider injection molding during design, and one supplier even quoted me for $20,000. So yeah, very glad for JLC!

Do you have any insight on how to take spectroscopy measurements at home on a <1k€ budget? And how to select an LED manufacturing supplier when CRI is often the only thing available on the datasheet?

Can't call it sun without gravitational confinement nuclear fusion either.

It's partly the photography but definitely the reality as well. Mine is only 0.03m², the DIYPerks parabolic reflector is 1.13m² :)

One of my goals was to explore the feasibility and scalability of a refraction based design.

> windowless room

Thankfully that's illegal where I live. I'm even tempted to suppress my own work here, because I don't want it to be used to unlock new levels of dystopic housing.

Just kidding, thanks for the feedback!

This is the reason I decided to use 4000K instead of the more physically accurate 5400K. I picked a less middle of the day color temperature and closer to evening time, basically.

I spent a lot of time looking at LED suppliers and there's quite a range of options in terms of color temperature. Some manufacturers even have 2 color temps on the same LED package, meaning you can sort of do what you describe by mixing two light sources at a distance less than a millimeter.

> Honestly I want to build something like this but my area of expertise is too far into the software domain, and not very far into the hardware or electrical engineering domain.

That was me last year before starting this project! You'd be surprised what you can achieve with time and effort!

That's a good point, but not much I can do about that. Such a DIY project is limited to off the shelf LED suppliers. It would be cool to do such a lamp with both the high CRI and some infrared, also for heating (infrared lamps are a thing after all).

Yes that's me! Happy to chat :)

Thanks! Feel free to reach out, I love to chat about this stuff. Contact info is on my website.

Thanks for the feedback. Everything I know about PCB design is self-taught. I'm a total beginner, so I'm sure there's a lot to improve!

There's only 2 routes per board, VCC and GND. I initially planned for SMD header pins that I didn't end up using, because soldering wire on bare pads was good enough. I also planned for 8 connection pads per PCB, but only used 2 to 4 in the final assembly. So yeah, lots of room for improvements in the PCB design! Definitely would need to spend some time on it for a higher power version 2.

Are you hiring? I'm looking for a job currently. Contact info is on my website :)

Turns out you can yeah! I'm not sure the quality would be anything useful for imaging applications though. But for lighting / non-imaging stuff go for it!

That's a great suggestion, thanks! I'll give it a try. It already works with dust quite well though, hehe.

I've given some thought to this, I agree it's one of the obvious improvements. I think with a refractive design like this one it's difficult, because of the way a lens works. Most other projects like this including commercial ones like CoeLux use a reflective design, and I believe that's partly for this reason.

I see. Yeah I could look at that but it needs to be re-implemented in pytorch to provide the backwards pass anyway. But the implementation might be a useful resource.

Thanks, I've fixed the links.

Honestly, the main challenge at this point feels like finding the time / resources to just work on everything I've got on the roadmap. I think I've got enough ideas for at least 1y of full time work, perhaps more.

Bezier splines are interesting. I actually had a working implementation a few weeks ago, so I know it will work. It's just that I refactored the internals of how surfaces work and didn't port the old bezier spline code yet, but it shoudln't be too difficult.

I haven't given too much thought about volume representation so far. Currently it's just a list of unconnected surfaces which seems sufficient for now.