Show posts

This section allows you to view all posts made by this member. Note that you can only see posts made in areas you currently have access to.

Show posts Menu

Messages - Samuel H

In theory, it's possible. In practice, Canon seems to be doing off-spec that makes this very, very difficult, unless you have insider info on what it is that Canon is doing off-spec.
Just so you get an idea of why everybody is going at you so harshly: what you're proposing is to design a camera from scratch, using a sensor salvaged from an existing camera, having to reverse-engineer all the communication protocols of the sensor. Designing a camera is hard enough as it is, even when you have all the documentation for the parts you're using. It's basically impossible if you don't have that documentation.
Again, my needs are my needs, and other people may have completely different needs, but I would be fine with just one 128GB card, it can hold 25 minutes of 1920x1080 24p video, and I don't think I've ever shot more than that in a half-day. So, just offload and make the backups at lunch, and I'm set.
And by what you posted it seems some buyers were a bit unlucky, but from what I've read around the web the Transcend 1000x cards are usually fast enough.

I obviously prefer SSD drives, but if the adapter costs $300 I'll probably just buy a CF instead. Not saying everyone will do that, just: keep it in mind that the price of CF cards is falling rapidly, and that brings down the price people will be willing to pay for the adapter.
I'm not saying there's no need for this. I'm saying it may not be as big as some might think, because yes, my needs are met by CF cards, and I'm guessing I'm not the only one. For me it would be a matter of cost, and that gap is closing down quite fast, with good 128GB cards already at $300 (so the gap is less than half of what it was 4 months ago).

SSDs are better and cheaper, but make sure you have a big-enough market before you get a mortgage to finance this. That's why I suggested a kickstarter campaign. Still problematic, though, if it turns out it's not in your hand to fulfill your promise, because of some technical impossibility.
My guess is that the sensor won't do more than 30 fps, no matter how fast your card is. If it does more, it is by lowering the resolution, so then you don't need faster storage either.

If the sensor can't send out more data than what a fast CF card can handle, then switching to an SSD drive is pointless. Except for cost.
I don't need speed. I'm perfectly fine with 1920x1080 RAW video, which tops at 83 MB/s. As long as I have that, I'm fine.
Of course, some people would disagree.
* it seems to me that it is not easy to estimate demand for this thing, so, a kickstarter would be a good idea, since it commits buyers before development takes place

* the bad side to this is that a possible outcome is that they find that it is not possible to use an SSD drive, because of some issue with the Canon card controller, and then the buyers are taking all the risk

* also, the price of CF cards that are fast enough for shooting RAW on the 5D3 has already fallen by 50% since this started (for 128GB, you used to need a $600 SanDisk card, now you can use a $300 Transcend card; no, I won't consider KomputerBay a viable option), making the adapter much less compelling; a year of development would make this even worse (my price for this gadget would be: the price difference between a 128GB CF card that is fast enough, and a 120GB SSD drive; and this difference is falling really fast).
^ You also need some logic to control the process. I.e. a processor, and some programming so it does something meaningful when you press the buttons. It would be cheaper if it was a mass-produced item...

The alternative is to buy a small notebook with loads of HDD space. Much more useful, but it weights a lot more.
For me, the main reason for using an SSD instead of a CF card was always price: if a 128GB 1000x card is $600 and a 180GB SSD is $100, well, I would be very happy if I could buy this adapter for $300. But with a good 128GB 1000x card at $300, well, it's not so attractive...
I abandoned development when I saw lots of real engineers already working on it. It seems they're stuck because the camera does some out-of-spec stuff, or otherwise requires something that a standard mechanical CF-to-IDE adapter plus IDE-to-SATA adapter won't do. If that's the case, the adapter would have to be more complicated, with a custom chip probably, and that makes it a lot more expensive and slow to develop.

In any case, with Transcend 128GB 1000x CF cards offering storage for 24 minutes of 1920x1080-24p RAW video for $300, my interest in this has fallen considerably...
The problem right now is not that the electrical interface is not fast enough (SATA3 is 600 MB/s, or about 7 times faster than we actually need), but that we can't fool the camera into thinking it has a standard CF card and at the same time translate the received data for the SATA controller.
^ That sounds unnecessarily complex, you can get the same result with the relatively dumb CF-to-SSD device that people are trying to design
(not saying we'll get 3.6K recording, just that anything that is possible with your concept should be possible, in principle, with the "simple" CF-to-SSD device too)

btw: I don't think I posted this, but I abandoned this project when I saw there were lots of more qualified people, even firms, working on it
^ A couple of very good finds there!!

Powering the adapter from the CF slot is actually a good thing: another possible reason for the camera not wanting to work with the adapter could be that it is trying to power a card, and it's finding either an incomplete circuit, or a null load.

The good news is, I now can put down the soldering iron and move on to something else. Man, these pins are small!
^ Yes, I know, but esskabel products are not easy to buy, and others are working on that front too. Right now, I'm just looking for "proof-of-concept" on a CF card reader.
No, just lots of patience, some ingenuity, and a couple of CF cards to trash. I know the most likely outcome is a useless adapter...
Quote from: pascalc on June 01, 2013, 05:57:39 PM
Hi Samuel, Can I ask you which CF extender did you choose ?
On mine, the link between the card and the exterior socket is a 40 pins ribbon. Some connectors are merged (surely the grounds ones). It works in a Lexar usb 3 card reader on my computer (with slow transfer rate) I don't understand why it doesn't on the camera.
I would like to try something with all the pins directly connected to the external socket but it seems to do not exists.

I'm trying to build my own ribbon adapter. Crazy, I know: the CF connectors are SO SMALL!
I bought a cheap cf-card-on-ide-bus adapter to have as an extra reference where I can see what's connected to what in a working product.
I still have a few parts missing, but I'm playing with what I've got. I've found that at least in some IDE-to-CF adapters there are a few extra connections beyond what this shows:
For example, in the one I've analyzed, in the CF connector, pins 8, 9, 10, 11, 12, 14, 15, 16, 17 and 25 are not "unconnected", but "tied to pins 1 and 50" (which then go to ground). Not the case for pin 24, though.
This may mean that there's something missing in that pdf, or, maybe more likely, that some parts are not actually made to exactly meet the full specification, which could explain the issues some here are having.
This says that those pins are used for the address bus; the CF-to-IDE connector described in the pdf would leave them unused, but if they're linked together and to ground the camera may freak out and shut down.
the problem may not be how David's converter sets the WB, but whether Resolve reads that metadata or just sets a default value

So, the image you labelled Resolve comes from a MOV file that is at least 5 times smaller than the stream of DNG files, and it plays back exactly how fast?

Finally a reason to get my paws on Resolve... At least to get an idea of what it can do, and what those limitations are (WB and curves).

I understand you can recode the DNGs as a CF-RAW mov file, edit them in Premiere (with real-time playback, choosing a low quality debayering method if needed), then export the whole project to be color-corrected and rendered in Resolve.
Seeing those results, this would be my preferred workflow for anything that doesn't have VFX.
My project is on hold, as I wait for some parts to arrive... grrrrr...
OK, so Resolve does a great job too, great to know! Thanks for the test.

The cineform sample has some color artifacts; was it debayered by the cineform software, or with the Resolve algorithm? If by CineForm, which debayering method did you use? The one I like best is Advanced Detail 1 (coincidentally, the one DANewman said was winning awards only 4 years ago), but if there's a way to use CineForm RAW compression and get an image that is as good as that Resolve sample, that's what I want!
Quote from: AndreasK on May 28, 2013, 10:20:40 PM
ehm Samuel wait, I did not run cineform on that test. I did CR2 (still image), RAW (via ACR) and H264. To be honest I don't remember the sharpening settings but I think it was just the default 25 with radius 1.0 - For the CR2 I used resize sharper in photoshop to resize to HD resolution.

Isn't CF advance detail 2 the best option for good sharpnes with no excessive artifacts David?

ooops, I don't know how I read that, I thought I was seeing H.264 vs CFR vs RAW

I guess I was surprised to see such a big difference between the downsampled CR2 and the video RAW, and I just assumed it would be something intermediate, like CFR.
I made a similar comparison back when this was not public yet, and in my memory results were really close. Looking at them again, well, closer than H.264, but not quite the same:
Quote from: AndreasK on May 28, 2013, 11:00:10 AM
Take a look at my comparsion pics Samuel. It may not be as sharp as ACR but no way it's as mushy as the H264 :)

edit: I'm dumb and what I wrote here was wrong
Quote from: DANewman on May 28, 2013, 06:51:54 PM
As the de-mosaic is not baked, I happy to hear suggestions on for alternative algorithms to add, making the CineForm RAW results even better. Currently you can use CineForm RAW (as .MOV) directly in Resolve, so you aren't limit to our de-mosaics today.
I'm afraid I don't have any particular method to suggest. The ones I've coded myself were relatively basic ones that would not improve over what Resolve already has. Now, if somebody finds out what ACR is doing...
This is a great option for fast-turnaround projects: if it runs at over 100 frames per second, it's basically just like converting the .raw file to a stream of .dng files! But smaller and with proper playback!

So, with something like Advanced Detail 1 debayering, how does detail look compared to the original H.264 video? CineForm RAW is clearly softer than an uncompressed DNG opened in ACR/AE (not because of the compression, but because of the awesome debayering that ACR uses, which CFR seemingly can't match). It should be somewhere between that soft ugly thing we used to love but don't want anymore, and the sparp, clean beauty of ACR