> "Expose for the shadows, develop for the highlights"
As a historical note, this practice has been mostly unworkable for a long time. I haven’t met anyone who does this.
The basic idea for this technique is to use a spot meter to measure both the dark and light parts of your scene, and adjust exposure and development time to map those to the correct tonal values. There some big reasons people don’t do this:
- You can’t change development for pictures in roll film,
Yeah. I’m not really a professional here but I would say sensitometry is something of an interest of mine. Take what I’m saying with the appropriate grain of salt.
Your camera has some sensor which records some range of values. The range is a lot wider than your monitor can reproduce and generally follows some curve up to a maximum value, where the highlights clip.
Log is something of a fantasy in that it represents an idealized, mathematical idea of what your camera does, but it does not really correspond to what the sensor does or what the monitor displays. You can convert your video to log, or from log, but the log data itself is only really useful as an intermediate step when post-processing. I don’t think you need to capture in log. I would just go for some “regular” wide like Canon’s Wide DR.
YMMV, I am a little old-school about these things and there is a lot of stuff that I prefer to do in-camera. (Be warned that while doing things in post gives you flexibility to undo your work and change your mind, there are a lot of things you can’t reasonably do in post… even things that seem basic like white balance are better to do with lighting or filters, and if you don’t understand the sensitometry then you can easily end up with footage that you can’t rescue in post.)
The one thing worth considering here is that the noise is intrinsically (short of a handy laser illuminated scene IIUC) proportional to the sqrt of the illumination level. So we should recored in sqrt for better coding efficiency/range :)
"there are a lot of things you can’t reasonably do in post"
I have my own list of these things.
Mine is a bit general, including for example the anti-aliasing filter, and polarizers. I'm curious about yours. Would you mind to share?
This is exactly the one thing I was thinking about. There are a lot of (image-related) things you can’t reasonably do in post, but that’s an especially important one.
I am even less of a professional and this whole field looks exactly like audiophile snake oil peddling. "log" makes total sense because codec cannot encode full precision sensor values (otherwise, just save that!) Any floating point is "log". As long as image carries color space and the mapping LUT, you can convert it into a linear space, which is only thing that makes sense for most operations. Instead, people are taking these "shmony super log8 plus ultra" files, and looking at them directly! they say "log looks flat", they adjust contrast and curves on it, ... How is that making any sense? What "look"? Why is reversing any such LUTs not done automatically? Super weird
log _was_ a compromise. It allowed one to do film scanning at the right level between precision and storage.
cineon, which is where you could argue that log came from, was a 10bit format to represent something that could have easily been 16bit. The compromise was that you kept the precision in the blacks for recovery, but with the ability to actually handle the files on hardware at the time. (bear in mind that realtime 10bit colour playback at 2/4k required a rack full of harddrives up until at least 2010 (see baselight 8 hardware))
I would hope there are 14 and 16 bit formats that come out of the camera that allow you to have some sort of linearish recording. Now that sensors are actually capable of capturing a dynamic range greater than top end film stock. However light isn't linear, so log is actually closer to how we see than linear is.
What ever it is, there are many levels of hidden LUTs to get the camera to record in a way that its meant to. what ever you are doing involves fiddling someone else LUTs
That part makes sense. But why then go and start applying operations that require linear colorspace (contrast, blur and sharpness, brightness, scaling and rotation etc etc) to data that is not in linear colorspace? Like, zero tutorials on "how to grade log" I have seen starting with "first we convert to linear colorspace such as Blah". What is this about? Because it looks totally wild and I assume I am missing something, the punchline.
that would be because lots of artists don't know how to do math. There was a 5 year period where every CG heavy film had obviously incorrect color if you knew what to look for because ACES 1.3 had a broken color space.
Writing English in the AI era is like going for an hour-long walk in the park, when previously it took you 15 minutes via a crowded subway, where you were stabbed in the neck by a raging lunatic while a nearby passenger blasted Coldplay at 100 dB.
It’s like walking in the park when everyone else is zipping past you on scooters, and there’s nobody going your speed for you to talk with or even smile at.
I think "previously" meant "previously, using AI," not "previously to AI." The comment is confusing because "the AI era" seems like the antecedent of "previously" but it isn't.
These days setting up a proper system on your build orchestrator of choice is pretty easy, cheaper, and you basically only lose a tiny bit of UI niceness if they report to github anyway
I work at Buddy. The CI product matters less than it looks here, most of that bad histroy is third-party actions pulled by a tag which someone else can move. Plain scripts as every step fix most of it on Buildkite, and also on GitHub itself, which is the cheapest option in my opinion
Just about anything else, tbh. Jenkins is the option I would go for, but that’s a reflection of my personal preferences for self-hosting. Self-hosting has gotten way easier since LLMs can debug things so well, so I’m self-hosting more than ever before even though I have less time available.
You don’t have to go Jenkins, it’s just that GitHub Actions are uniquely bad and I would avoid them.
If you are relying on the free version of GitHub actions for small projects, I would consider just running something on a NUC or Pi at home.
A Pi is not going to make your Windows and macOS builds. It's also going to have a miserable time running the Rust compiler on your five hundred dependencies.
Here’s my take: if you don’t personally have a Mac, you should not be making Mac builds of your software.
Same for Windows.
I may be a little old-school about this. That’s fine. I think most people out there who are trying to support multiple platforms with their code should be going one of three directions—(1) some kind of web app, (2) an open-source tool where end-users compile it or run it in their Python interpreter or Node.js or whatever, or (3) a pre-built binary for systems they personally run at home.
I also am not writing out, like, an essay on how to replace GitHub actions for every possible use case that people have GitHub actions for. I am just providing some suggestions for one possible direction. If you get off GitHub actions, you are going to end up with something that is missing some of the features that GitHub actions, maybe some features that you like. The same is true in the reverse—if you switch to GitHub actions, you will obviously lose features that were present in whatever other system you had before. GitHub actions is, after all, completely terrible and switching to it would just be a kind of disaster in a bunch of different ways.
I don’t like to put this kind of wall-of-text disclaimer in comments on HN, and hopefully y’all can figure this stuff out without the wall-of-text disclaimer. It would be kind of… absurd… perhaps, to expect some random commentator on Hacker News to give you exactly the answer that satisfies your needs, unless you put those needs in a question.
Instead I limited myself to two suggestions. If you don’t like them, then figure something else out yourself. You will not have trouble making something better than GitHub actions.
> if you don’t personally have a Mac, you should not be making Mac builds of your software
Apple seem to believe this extremely strongly and keep finding new ways to prevent you doing it.
I basically agree, and at work we have a Jenkins master + cluster of build machines. But this does incur substantial sysadmin overhead even if we're using Ansible and all the Windows ones are VMs. (Mac, again, hates being either virtualized or stuck in a rack, and will occasionally insist on someone using the physical display and keyboard on the actual machine to perform updates)
A 'should' with no explanation? What if I think that I 'should' release, for my tool I want people to use, binaries that run on their computer instead of just mine? Why 'should' this be a web-app? (What's 'old-school' about that?)
GHA over Jenkins is insane. Every time I’ve touched GHA, I’ve just been shocked with how bad the ecosystem is.
It feels like, I don’t know. You go to somebody’s house and they have one plate, one cup, one spork. They’re living in a 30th floor walk-up, but one of the windows is missing so there’s a giant hole in the wall.
That’s kind of the subjective experience I get when I see people using GitHub actions.
I'm not saying GHA is much better, but Jenkins is literally swiss cheese, it has vulnerabilities every day of the week. Even just as an internal service not exposed to the internet it's dangerous.
This was my second hackernews submit [0]! So, not even https in 2014 :)
When I ran that server, outlook.com was also the only one that would blackhole my email for no known reason. Support was unhelpful and also could not point to any reason. I've since given up, but for a short time it was glorious, email to family would not leave the lan and we could send attachments of arbitrary size.
Your specifics might be accurate, but the broader point is not. WhatsApp. In some parts of the world, it is de facto required for a business to have WhatsApp, because "everybody has it". Yes, you would message the hotel front desk from your room, using WhatsApp to ask for a new towel. I have.
You adopted it the moment it was announced? Really?
Is there a risk some people on this thread are mixing up DKIM and DMARC?
DKIM = proving that an email was sent from a specific domain using cryptography. Good for spam fighting and deliverability.
DMARC = telling receivers how to handle mail that fails authentication.
DMARC is primarily useful for large, complex organizations where email admins may not be aware of everyone inside the org who is sending mail that advertises the domain in the From header, or organizations that might be targeted by impersonation attacks (e.g. phishing). Being able to tell recipients "yes we absolutely have our act together so feel free to reject or spamfolder unauthenticated email" is great for killing phishing campaigns, and being able to request reports is great for finding that one sales org that signed up for a guerrilla marketing SaaS that's impersonating them.
But for small personal domains the value isn't that high.
I added DMARC and DKIM together, around the same time as some particular life events 10 years ago. Just now, I cross-checked the information against the commits in my Git repos.
Back then, I didn’t use Terraform or anything like that, but I had text files with logs that described the DNS setup in detail so I could fix things. By grepping through the logs, I’m able to put a definitive bound on when I added DMARC—it was no later than June 2016, over 10 years ago.
At the time I thought, “Obviously, we should all be setting this up, because recipients will start rejecting mail from insecure domains, if they don’t already.” This was also back when I self-hosted email. I eventually stopped doing that because I thought it was too time-consuming.
>DMARC is primarily useful for large, complex organizations
>But for small personal domains the value isn't that high.
I don't understand the resistance. It benefits the entire mail ecosystem, not just mass senders.
SPF used to be niche thing, then everyone realised it's a good idea and it became more-or-less mandatory. Same with DKIM. Now it has happened to DMARC as well.
No DMARC is about telling receivers which of SPF/DKIM are valid, what to do with it when it fails, and where to send a report to. Oh look.. like Mike said.
DMARC is primarily about "alignment" (switching SPF from caring about the envelope sender to instead using the from header). Yes, it does also tangentially address those other points, but the primary reason it was adopted was to prevent spoofing of from headers (where in original spf that was a non-goal)
I used to do floppy disk data recovery as a service for students at college. Roughly speaking, we helped one student out every three hours, out of 2,500 students, during busy hours. Based on the rates and hours, I think it would come out to something like 25% of the student body, each year, coming to us with a data loss incident involving floppy disks.
YMMV. I’m sure a lot of people were shoving these things into backpacks, and this was post-2000, so the quality of disks for sale was poorer than it was in the 90s and 80s.
I’ve also (since then, for myself) repaired drives. The drives fail because they’re mechanical and have motors, and anything that moves or has a motor in it will need some kind of service. Servicing a floppy drive requires specialized equipment which is hard to find (or requires a lot of patience—it’s possible to repair a floppy drive with just a screwdriver, but it’s not fun; guess how I know?)
HD floppies were much worse, both 1.44MB and 1.2MB. They are just much more susceptible to defects from wear and tear because of the shrunken sector size. DD 3.5" wasn't as bad.
I am guessing a combination of tighter manufacturing tolerances, smaller sector sizes, and manufacturers over time cheaping out would contribute to decreased reliability of 3.5" disks. People also, as far as I remember, universally used sleeves for 5.25" disks, but almost never had cases for 3.5" disks.
On the other hand, 3.5" disks have higher magnetic coercivity, which makes them less susceptible to data loss from stray magnetic fields (e.g. nearby motors).
I don't think you needed cases for them though since they covered the magnetic disk with their own sliding door while on 5.25" disks they were exposed to the elements. I once spilled jelly all over a 5.25" disk and took it out of the sleeve, washed it off in the sink, and put it in a new sleeve and it still worked :)
People who can figure out words from radicals probably have >500 word vocabularies to begin with. I’d say if you started learning on some natural progression, you would start learning how to figure things out from radicals once you knew somewhere around 300 kanji (which would mean you have a vocabulary of 1,500 words, ish).
Before that point, you are probably going to have too much trouble generalizing.
It’s just a choice of which key to use. I don’t think it was some decision made with this intent, I think it is just the unexpected result of a decision.
As a historical note, this practice has been mostly unworkable for a long time. I haven’t met anyone who does this.
The basic idea for this technique is to use a spot meter to measure both the dark and light parts of your scene, and adjust exposure and development time to map those to the correct tonal values. There some big reasons people don’t do this:
- You can’t change development for pictures in roll film,
- We have good variable contrast paper,
- It’s harder to do in color.
reply