Welcome to our community

Be a part of something great, join today!

Log in
  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

October 30th Announcement Feedback

Status
Not open for further replies.
Peter Majtan said:
John Sagud (from RED) have said this:

"Scarlet will offer the 35 mm still camera equivalence of a 28-224 mm zoom lens. (This would be described as an 18-144 mm on an S35 sized sensor, such as the RED One and 7.2-57.6 mm on a 2/3" sensor). There are also planned auxiliary wide and tele adaptors for this fixed lens."

Lets see if the 30th November announcement brings any changes...

:devil: Peter
Click to expand...

s35 18mm fov is equivalent to 2/3 7.2mm fov ?
So thats a 2.5 multiplier for fov conversions between the two?
 
Just as you let us know that the Oct 30th announcement would take place around 3pm PST, could you let us know when you expect to release the info on the next 30th announcement?

That way I can make some plans on our next podcast. :emote_popcorn:
 
J Davis said:
s35 18mm fov is equivalent to 2/3 7.2mm fov ?
So thats a 2.5 multiplier for fov conversions between the two?
Click to expand...



RedOne's S35 has a diagonal size of ~28mm, 2/3 has ~11.4mm, that gives a relative crop factor of 2.45 =)

Jojo

PS: Relative crop between the new 30mmx15mm S35 sensor and the 2/3 is ~2.93
 
Jeff Coatney said:
The prototype they showed us was a pretty sophisticated sub-assembly. My guess is, they're pretty far downstream on this. So yes, its a big deal at this point.

You should withhold judgement until you see the total ergonomic impact. Maybe having extra rings on the lens barrel would require the shooter to have a third arm to pull focus. If you want a first AC at your side the rest of your life, buy the model with interchangeable lenses.

Of course I'm sure they'd scrap it in a minute and put Scarlet back in the development cycle for another year or two if enough people squawk. Is this issue a deal-breaker for you or are you willing to adapt?
Click to expand...

My initial fear, has turned into peak interest. We've been told we can assign Iris and zoom to the handle, in a way I think that could be better then on the side of the lens. Our hands can stay firmly on the camera in hand held without having to move around. Now that I am satisfied It can do what I know I need, I am starting to think about the new possibilities it will add to my shooting. Program rack focus and zooms seem like a cool think to explore.

I this camera is catering more to the people who won't have a crew, so it's great they have thought of this. Be interesting now to see how it all works.
 
Peter Majtan said:
John Sagud (from RED) have said this:

"Scarlet will offer the 35 mm still camera equivalence of a 28-224 mm zoom lens. (This would be described as an 18-144 mm on an S35 sized sensor, such as the RED One and 7.2-57.6 mm on a 2/3" sensor). There are also planned auxiliary wide and tele adaptors for this fixed lens."

Lets see if the 30th November announcement brings any changes...

:devil: Peter
Click to expand...

I think that is pretty wide. It sounds very nice. It would of course have been nice with a little higher number in the other end, but then again, maybe that only gives us a lens with better quality, and with less light loss.
 
When will we be getting ball park dates/prices on the Elenses?

Are we talking $3k, $9k, $15k lenses?

Is the 25-100 going to be about $10k like the 18-85 s35 pro?
 
Martin Weiss said:
May one dream of face recognition?
Click to expand...

This is probably the best solution to passive-focus systems. I worked on a face-recognition project some years ago in cooperation with Los Alamos Nat. Lab. At this point in face recognition development, there are many solutions yet all of them are security based denial-of-access platforms. Finding terrorists in a crowd, identity confirmation and so on.

Thinking about this for awhile. I personally had in development (initial exploration) a passive-focus system that would have dynamically updated distancing and error correction at several thousand samples per second. The problem I ran into is primarily the cost of making an attachable drive mechanism capable of matching the vast number of lenses in the marketplace. The mechanical interface that actually moves the lens' focus ring.

I got very excited by what Curt Von Badinski at Viewfactor is doing- he has a fantastic design for this component, but he has been undermined by poor vendor support. However, in the case of the fixed lens Scarlet, now that I've seen the lens assembly, it would make sense to incorporate a software based solution.

Dynamic Bio-metric tracking of faces or other selectable or user-assignable reference points could drive the focus motors on a platform like Scarlet. I'm sure this has been explored at Red and may even be successfully integrated into Scarlet's design. If not, this could be implemented in a firmware update, depending on how much surplus computational capacity is in the system.

Traditionally with face-recognition, you have a very-high computational intensity that uses the total amount of processing capacity in a very small time-scale-- milliseconds. But previous technologies have been built on neural-network processing that builds a database of decision trees linked to a fixed database, which is vast. This accounts for the limited implementation at strategically important high vulnerability sites like airports, military bases, federal reservations and corporations with intellectual property critical for national security.

But in a passive-focusing system, we're just talking about a data base of a dozen or so bio-metric profiles that can be user-assignable and stored in ram or flash memory. The bulk of the computational resources is used for image processing. I have no idea how big the processing capabilities are on the board stack in question, but it would seem its at least possible to piggyback a system that talks to the lens using existing computational bandwidth. I realize we're slogging through a vast amount of resolution coming off the sensor, and Red Code is using the lion's share of it, but it would be a worthwhile exercise to determine the passive focusing system's computational needs and then integrate it into the existing system.

Graeme would know how much room there is on the wafer with Red Code running full-tilt, so to speak. If there's no more room, a daughtercard could handle the task.

Face-recognition, which takes a single sample image and tries to find a solution from a database with over five hundred thousand possible matches in it, in as fast as a second or two. Such a passive-focusing system needs to take a multiple of samples based on the current frame-rate and compare differences in those samples, but it would have to do it in a fraction of a second and interpret that data into executed lens motor commands within the the time-relevance of the shot (near real-time). Preferably, it should build a probability profile that actually predicts focus based on sampling data that determines the subject's trajectory on an axis relative to the lens, regardless of the lens orientation.

Then you've got something.
 
Jeff Coatney said:
This is probably the best solution to passive-focus systems. I worked on a face-recognition project some years ago in cooperation with Los Alamos Nat. Lab. At this point in face recognition development, there are many solutions yet all of them are security based denial-of-access platforms. Finding terrorists in a crowd, identity confirmation and so on.

Thinking about this for awhile. I personally had in development (initial exploration) a passive-focus system that would have dynamically updated distancing and error correction at several thousand samples per second. The problem I ran into is primarily the cost of making an attachable drive mechanism capable of matching the vast number of lenses in the marketplace. The mechanical interface that actually moves the lens' focus ring.

I got very excited by what Curt Von Badinski at Viewfactor is doing- he has a fantastic design for this component, but he has been undermined by poor vendor support. However, in the case of the fixed lens Scarlet, now that I've seen the lens assembly, it would make sense to incorporate a software based solution.

Dynamic Bio-metric tracking of faces or other selectable or user-assignable reference points could drive the focus motors on a platform like Scarlet. I'm sure this has been explored at Red and may even be successfully integrated into Scarlet's design. If not, this could be implemented in a firmware update, depending on how much surplus computational capacity is in the system.

Traditionally with face-recognition, you have a very-high computational intensity that uses the total amount of processing capacity in a very small time-scale-- milliseconds. But previous technologies have been built on neural-network processing that builds a database of decision trees linked to a fixed database, which is vast. This accounts for the limited implementation at strategically important high vulnerability sites like airports, military bases, federal reservations and corporations with intellectual property critical for national security.

But in a passive-focusing system, we're just talking about a data base of a dozen or so bio-metric profiles that can be user-assignable and stored in ram or flash memory. The bulk of the computational resources is used for image processing. I have no idea how big the processing capabilities are on the board stack in question, but it would seem its at least possible to piggyback a system that talks to the lens using existing computational bandwidth. I realize we're slogging through a vast amount of resolution coming off the sensor, and Red Code is using the lion's share of it, but it would be a worthwhile exercise to determine the passive focusing system's computational needs and then integrate it into the existing system.

Graeme would know how much room there is on the wafer with Red Code running full-tilt, so to speak. If there's no more room, a daughtercard could handle the task.

Face-recognition, which takes a single sample image and tries to find a solution from a database with over five hundred thousand possible matches in it, in as fast as a second or two. Such a passive-focusing system needs to take a multiple of samples based on the current frame-rate and compare differences in those samples, but it would have to do it in a fraction of a second and interpret that data into executed lens motor commands within the the time-relevance of the shot (near real-time). Preferably, it should build a probability profile that actually predicts focus based on sampling data that determines the subject's trajectory on an axis relative to the lens, regardless of the lens orientation.

Then you've got something.
Click to expand...


Definitely some great thoughts here Jeff, what if in order to shorten times of execution, a prerecorded small number of faces (when possible) such as in Film projects, was to be uploaded in a macro memory chip (on board).

Would this allow for an almost immediate and realtime execution of Focus to Face recognition do to the very small number of calculations as the faces to be focused are already in the system.

Would this be a possibility, same as in the security systems you talking about, as the authorized people will have their faces prerecorded and recognized y the security systems as to grant them access.

Would this be possible to be applied on the Electronic component which would control the Focus motor drive on the Lens?

Just a tough..
 
Only problem with facial recognition is what happens when someone turns around or something passes in front of their face? Does the track go crazy and introduce really bad wild zooms?
 
Pietro Impagliazzo said:
If these are lenses to compete with Canon/Nikon electronic lenses don't you think these prices are kinda high?

It would be reasonable to expect the most expensive to be around $3K...

Right?

http://www.bhphotovideo.com/c/searc...94951197+4294951196+4294951195&Ns=p_PRICE_2|1
Click to expand...

Perhaps but they seem to be selling these things as true cine-glass with only minor breathing etc..

Personally I'm hoping for a $1k-$3k per lens price point similar to canon but I'm unsure what market they're targeting with the e-lenses. And it will dictate whether or not I buy a set to just have on me at all times or they're a rental item and I rely on a less expensive still set for day-to-day shooting.
 
1 to 3k seems very low for e-lenses that will have smooth barrels, minimal breathing, and OLED screens on them, LDS, work probably better than the others with the new touch screen system.

Instead of comparing them to Canon or Nikon. Zeiss, Cooke, Fujinon would be a better comparison.

Even when Zeiss took the ZF lenses, gave them a smooth barrel, PL, and a few xtras the price jumped 6x fold.

10K for the 18-85. I would guess 13K, and that would be a great deal
 
Face-recognition, which takes a single sample image and tries to find a solution from a database with over five hundred thousand possible matches in it, in as fast as a second or two.
Click to expand...

The cameras I have seen are using face detection, not recognition. (I consider them two different problems to solve) It would be difficult to squeeze a Feret database into a camera. :) Its all about training anyway. PittPatt has a pretty decent detection algorithm. It takes about 2.5 seconds for a single frame of 1080p video. This is on a high end multi core box. OpenCV is pretty poor by default. With enough training, it might improve.
 
Vladimir Eugene said:
1 to 3k seems very low for e-lenses that will have smooth barrels, minimal breathing, and OLED screens on them, LDS, work probably better than the others with the new touch screen system.

Instead of comparing them to Canon or Nikon. Zeiss, Cooke, Fujinon would be a better comparison.

Even when Zeiss took the ZF lenses, gave them a smooth barrel, PL, and a few xtras the price jumped 6x fold.

10K for the 18-85. I would guess 13K, and that would be a great deal
Click to expand...

But you're talking about the Zeiss Compact Primes, smooth mechanics, primes, not electronic and expensive (relatively). Whole different game.

Electronic zooms usually are between that price point... If RED decides to sell them at another price point, well, that's their choice, not sure how well it will sell.

Doesn't RED like to price stuff agressively?

Gavin Greenwalt said:
Perhaps but they seem to be selling these things as true cine-glass with only minor breathing etc..

Personally I'm hoping for a $1k-$3k per lens price point similar to canon but I'm unsure what market they're targeting with the e-lenses. And it will dictate whether or not I buy a set to just have on me at all times or they're a rental item and I rely on a less expensive still set for day-to-day shooting.
Click to expand...

Yeah... Let's wait and see.
 
Stacey Spears said:
The cameras I have seen are using face detection, not recognition. (I consider them two different problems to solve) It would be difficult to squeeze a Feret database into a camera. :) Its all about training anyway. PittPatt has a pretty decent detection algorithm. It takes about 2.5 seconds for a single frame of 1080p video. This is on a high end multi core box. OpenCV is pretty poor by default. With enough training, it might improve.
Click to expand...

No, no. My point is a FERET-like database is not required. In fact, you could perform the tasking requirements without "Face Recognition" as most people understand the term. You need a set of trackable reference points- they can be bio-metrically created if this technology is within reach, but not necessarily so.
 
Gavin Greenwalt said:
Perhaps but they seem to be selling these things as true cine-glass with only minor breathing etc..

Personally I'm hoping for a $1k-$3k per lens price point similar to canon but I'm unsure what market they're targeting with the e-lenses.
Click to expand...

All things with RED are naturally subject to change, but the features projected to be on the new RED Electronic Lenses have their pedigree and history in 35mm still lens industry, and to a lesser extent in the EFP/ENG lens industry. Features like auto focus, auto iris, image stabilization, etc. are 35mm still lens technology. AF and AI are also features on certain EFP cameras and lenses. None of those features (AF, AI, IS) were born, are used in, or are in demand in the cine industry.

There are many 35mm still zooms which breathe little, hold focus through a zoom, don't have image shift, etc.

In describing the new RE lenses Jarred compared them to Canon and Nikon - not Zeiss or Cooke:


"you will always be limited to the speed of the motors in any given lens. For some people that use Canon and Nikon lenses they are ok with that tradeoff..

The RED Electronic FF lenses are designed with motion tracking and motion recording as a foundation however, so the performance of those lenses in both motor speed, lens noise, breathing, etc. will far exceed ( we believe ) other electronic lenses on the market."

Jarred
10-31-09
http://205.234.135.241/forum/showthread.php?t=37010&page=40


These new cameras are positioned as DSMC cameras - and I believe in technology and utility the new RE lenses are aimed squarely at being DSMC lenses - a pefect match with the cameras to unleash the broad utility the DSMC cameras themselves enable. There also is no shortage of big, heavy, manual, expensive cine lenses in the marketplace. IMO the new RE lenses aren't targeted at competing with that market per se. Again, they're aimed at the combined DSMC and hybrid non-hardlined EFP markets.

These improvements in the RE lenses over usual 35mm still electronic lenses will naturally come at a premium. I don't expect the new RE lenses to be priced as low as the current premuim 35mm still lenses, but I certainly don't expect them to be priced anywhere near the level of cine lenses.

Interesting times ahead...I think a lot of people are going to be re-thinking their equipment setups and camera techniques for what they shoot.
 
Gavin Greenwalt said:
Only problem with facial recognition is what happens when someone turns around or something passes in front of their face? Does the track go crazy and introduce really bad wild zooms?
Click to expand...

You can limit the amount of correction so the wild zooms won't happen, but the problem you describe could pose a hurdle. I have a solution to this I've been working on.
 
a piece of windows/mac software can handle the import of photos of objects or people to be tracked, set priorities and assign parameters like "hold focus when out of frame" and ultimately create tracking points for the camera to use to maintain focus (usually high contrast)
Then the output of the computer generated tracking references can be fed to the camera to use while shooting.

A few photos under different lighting conditions and angles would be required to create reliable tracking references.

I think this subject requires it's own thread so let's take it to a place where we can explore it further.
Let's try to maintain consistency in this thread otherwise people will find it hard to sift through 1500+ posts...
 
Status
Not open for further replies.
Back
Top