RAW vs JPG

Regardless of how AI affects our sex lives - what a tangent this is - AI in cameras is great. Even my old M240 does a great job at aperture priority auto everything. Call it a "point and shoot" if you want. It makes good images of what I point it at. Remember, I take photos to please myself. If they please others, too, that is wonderful.

Now about that AI strumpet, . . . LOL
 
Hmm. I don't know that I can call what an aperture priority auto-exposure system does—or even a matrix metering aperture priority auto-exposure system—"AI". That seems a stretch way way too far.

The reason is that the hallmark of that which can legitimately called "AI" is adaptive learning. No camera metering system does *any* adaptive learning ... They simply take inputs, combine them according to a set a priori schema created by the individuals who wrote the controlling program, and set the exposure to the value that results. There's no learning or adapting involved at all.

I guess an "AI strumpet" can be programmed in the same way if the naïvete of programmer and audience is sufficiently adolescent... 😈
... but we're digressing well beneath the usual subterranean gutter. 🤣

At the end of the day, it is amusing and perhaps a little saddening to think that a bunch of knowledgeable cameraphiles could not appreciate the differences or advantages between a photo captured as raw data vs a photo rendered out of the camera as a JPEG.

G
 
Last edited:
"At the end of the day, it is amusing and perhaps a little saddening to think that a bunch of knowledgeable cameraphiles could not appreciate the differences or advantages between a photo captured as raw data vs a photo rendered out of the camera as a JPEG."

Sadly, that statement works both ways...
 
Time for another cat picture! 🙂​
DSCF2929.jpg

Fujifilm X-H1
Fujinon XF 56mm f1.2
My back porch - Yokosuka, Japan​
 
boojum said:
Regardless of how AI affects our sex lives - what a tangent this is - AI in cameras is great. Even my old M240 does a great job at aperture priority auto everything. Call it a "point and shoot" if you want. It makes good images of what I point it at. Remember, I take photos to please myself. If they please others, too, that is wonderful.

Now about that AI strumpet, . . . LOL
Click to expand...

Your first paragraph rates an explanation. Please!! I am all ears, eyes and, well, most of the rest of me...

Two programmers I know have given me the best explanation of Ai I've had so far.

As they told it - most phone call centers with impossibly long wait times and inefficient poorly-trained staff basically ignoring the client's complaint or query in favor of reading preset templates, will become an irritant of the past.

Instead, all calls will be answered almost immediately, however many are phoning in at the same time, by the call center's main computer system. This means no waiting, or almost none.

After a given number of punching numbers for Y/N responses or to select the nature of one's call, a computer will then take over and deal with the query. Anything out of the usual will be efficiently and quickly referred to a (human) specialist client services person who will respond right away to your call or (again via the AI system) book a return call, usually within the hour.

One will be spared the agony of having to listen to horrible tinned music with agency/corporate come-on promos or "we value your call" sh*tcrap every two or three minutes.

Sure, it will mean massive unemployment to the brigades of high school graduates who hit the job market untrained for almost anything requiring skill or for that matter more than basic intelligence. But as one programmer said, we will all be better off for it. Universal income will take care of the jobless. Most are young and live with their parents anyway. Fewer holiday trips to Bali or new SUVs with credit signed for by daddy will result, but to some of us that isn't a bad thing. Okay, call me cynical. I calls me realistic...

As for the sex part, well, I await a response. This can be summed up well by a comment made twenty-odd years ago by a co-worker. Computer sex wasn't his thing, he opined, the disk drives (all PCs had them back then, remember?) were, as he put it, "inadequate" for the purpose.
 
DownUnder said:
Your first paragraph rates an explanation. Please!! I am all ears, eyes and, well, most of the rest of me...

Two programmers I know have given me the best explanation of Ai I've had so far.

As they told it - most phone call centers with impossibly long wait times and inefficient poorly-trained staff basically ignoring the client's complaint or query in favor of reading preset templates, will become an irritant of the past.

Instead, all calls will be answered almost immediately, however many are phoning in at the same time, by the call center's main computer system. This means no waiting, or almost none.

After a given number of punching numbers for Y/N responses or to select the nature of one's call, a computer will then take over and deal with the query. Anything out of the usual will be efficiently and quickly referred to a (human) specialist client services person who will respond right away to your call or (again via the AI system) book a return call, usually within the hour.

One will be spared the agony of having to listen to horrible tinned music with agency/corporate come-on promos or "we value your call" sh*tcrap every two or three minutes.

Sure, it will mean massive unemployment to the brigades of high school graduates who hit the job market untrained for almost anything requiring skill or for that matter more than basic intelligence. But as one programmer said, we will all be better off for it. Universal income will take care of the jobless. Most are young and live with their parents anyway. Fewer holiday trips to Bali or new SUVs with credit signed for by daddy will result, but to some of us that isn't a bad thing. Okay, call me cynical. I calls me realistic...

As for the sex part, well, I await a response. This can be summed up well by a comment made twenty-odd years ago by a co-worker. Computer sex wasn't his thing, he opined, the disk drives (all PCs had them back then, remember?) were, as he put it, "inadequate" for the purpose.
Click to expand...

AI is as broad as computers are. From incredibly simple up to what is used to research fusion, and worse. I see it in its literal sense, artificial intelligence as opposed to native, natural intelligence, you know, that stuff we can have. On that basis what the software does in a camera is artificial intelligence. The Wikipedia link covers this better. As for telecommunications, my last job was with Pacific@Bell. We had one large customer who had a lot of incoming public calls to get a ring at his end before the caller heard a ring. The perceived effect for the caller was the called party picked up the phone before it rang. This was over 30 years ago. It spooked some callers LOL This was crude. You know that P@B has moved along.

I have three old Leicas and despite being old they churn out good images. The Pixii, newer, does with an APS-C sensor good images of great detail. And the X2D does magic. The Hasselblad engineers have created a great image, JPG or 3FR (RAW). I cannot speak as to how much like it is like the current rage of AI like in ChatGPT. But I can tell you it does its sole task of making images very well.
 
This is certainly going off topic...

25+ years ago my wife was using Neural Networks for object recognition in images. One of the first to use the Wavelet transform for object recognition rather than just for compressing an image. For image compression- wavelets have some superior characteristics compared to Cosine transforms. BUT- JPEG-2000 never really caught on.

As far as Raw vs Jpeg- I leave the Nikon Df and Z5 to shoot both. The M9, M8, and M Monochrom- uncompressed DNG only.

AI is over-hyped, over-promised, and over-rated. It is the next big bubble that is about to burst. It's a sink-hole for venture capital and investors. It makes for a decent search engine and it can turn out some interesting images. AI vs NS- NS wins everytime.
 
According to Nikon the F5 used a neural network as part of its 3d color matrix metering and used the information from the sensors to compare against 30,000+ scenes stored in the camera to choose the correct exposure.
 
I never trusted Matrix Metering, always set my Nikons to use center-weighted metering. That way I could visualize what the camera was doing and dial in a fudge-factor for what I wanted, using exposure compensation. No way to visualize what matrix metering was going to pick as the subject. I spent the 1980s developing image processing algorithms for recognizing and tracking objects.
 
mapgraphs said:
"At the end of the day, it is amusing and perhaps a little saddening to think that a bunch of knowledgeable cameraphiles could not appreciate the differences or advantages between a photo captured as raw data vs a photo rendered out of the camera as a JPEG."

Sadly, that statement works both ways...
Click to expand...
Which are the two ways that statement works? In other words, what are you implying?

G
 
shawn said:
According to Nikon the F5 used a neural network as part of its 3d color matrix metering and used the information from the sensors to compare against 30,000+ scenes stored in the camera to choose the correct exposure.
Click to expand...

Hmm. The F5 debuted in 1996. It's hard to believe that Nikon could have implemented a neural network into the rather simplistic, low power/low memory processor of the F5 given the technology of that date, and if they had, I would have expected to see it ballyhooed on their historical page about the F5 ... Debut of the Nikon F5 ... But there's no mention of a neural network there.

I do see an interesting discussion of "Automatic white balancing via illuminant scoring autoexposure by neural network mapping" which describes a patent held by TI in which the Nikon F5 is briefly mentioned in section [0021] as example of a neural network " ... trained on 30,000 images to relate exposure value to parameter values of the following image parameters ..." but it's a little unclear to me that the neural network is active in the camera or whether it is trained in the lab and then used to create a mapping library for the camera to use as a static data set for comparison purposes in ROM.

Interesting stuff, however.

I think I'll make a salad for dinner. Should I used raw carrots or cooked carrots? hmm hmm hmm ... 🤣

G
 
Last edited:
Godfrey said:
Hmm. The F5 debuted in 1996. It's hard to believe that Nikon could have implemented a neural network into the rather simplistic, low power processor of the F5 given the technology of that date, and if they had, I would have expected to see it ballyhooed on their historical page about the F5 ... Debut of the Nikon F5 ... But there's no mention of a neural network there. Where did you see this reference?

G
Click to expand...
On the page you linked to in the schematic of the metering system.
 
Godfrey said:
Right. But is it a map loaded statically into the camera for comparator purposes, or an active neural network actively developing the map as data is acquired?
Click to expand...
Per the schematic the cameras sensors are feeding a neural network to determine exposure.

Nikon has several patents from before the F5s time regarding using neural networks to drive AF tracking and to determine proper exposure from borderline cases from reference images. The exposure one makes it sound like a bit of a mixture of a neural network with simplification to reduce the in camera processor demands for borderline cases. I didn't dig too deeply into these though.

US5266984A - Exposure calculating apparatus for camera - Google Patents

US5864693A - Movement control method and chaotic information processing unit using chaotic neural network, and group movement control method - Google Patents

US5732288A - Auto-focusing device for camera - Google Patents
 
shawn said:
Per the schematic the cameras sensors are feeding a neural network to determine exposure.

Nikon has several patents from before the F5s time regarding using neural networks to drive AF tracking and to determine proper exposure from borderline cases from reference images. The exposure one makes it sound like a bit of a mixture of a neural network with simplification to reduce the in camera processor demands for borderline cases. I didn't dig too deeply into these though.

US5266984A - Exposure calculating apparatus for camera - Google Patents

US5864693A - Movement control method and chaotic information processing unit using chaotic neural network, and group movement control method - Google Patents

US5732288A - Auto-focusing device for camera - Google Patents
Click to expand...
Very interesting, I'll have to read through them. thx!
Of course, it's all way way far afield from the original question of the value of raw vs JPEG capture.

Little point to more discussion on this thread. And I just scanned 36 Minox 8x11 negatives ... I should get busy rendering them. 😀

G
 
You must log in or register to reply here.
Back
Top Bottom