Auto focus is 'Hot'
- Thread starter morgan_moore
- Start date
morgan_moore
Major Contributor
This? http://www.newsshooter.com/2014/04/...-focus-system-making-the-impossible-possible/
I spose it could be handy for drama where you can 'wire' the talent.
Thing is if you stick it in their jacket the eyes will be 6inches 'fore of the focus plane.
Also the chess vid was horrid, but I guess the focus speed will become controllable.
--
Personally I think 'focus 5%' is the way forward.
The puller put the focus at 30inches, the tech decides if it should be at 31 or 29 inches - going the last 5%
S
I spose it could be handy for drama where you can 'wire' the talent.
Thing is if you stick it in their jacket the eyes will be 6inches 'fore of the focus plane.
Also the chess vid was horrid, but I guess the focus speed will become controllable.
--
Personally I think 'focus 5%' is the way forward.
The puller put the focus at 30inches, the tech decides if it should be at 31 or 29 inches - going the last 5%
S
morgan_moore
Major Contributor
My POV on the history of focus.
I first shot 4k raw in 2003 - (http://www.luminous-landscape.com/reviews/cameras/kodak-dcs.shtml)
I very quickly realised that even on 6.45 ground glass for one frame on a static subject I could not close the last 5% of focus accuracy and shortly moved to AF for all still photography (coming from the standpoint of AF is for wimps)
Personally I think the 4k camera makers have completely failed to bring us tech that can lead us to resolve true 4k due to lack of focus aids/Af
I also think that future AF could have a big impact in the value of cinelenses and bodies like the aleza/F55 part of the reason Ive never jumped to any really big ticket cine kit as an owner op.
I predict that AF will sneak onto movie sets throught the back door, first used on Movis then for telephoto running shots then.. then..
S
I first shot 4k raw in 2003 - (http://www.luminous-landscape.com/reviews/cameras/kodak-dcs.shtml)
I very quickly realised that even on 6.45 ground glass for one frame on a static subject I could not close the last 5% of focus accuracy and shortly moved to AF for all still photography (coming from the standpoint of AF is for wimps)
Personally I think the 4k camera makers have completely failed to bring us tech that can lead us to resolve true 4k due to lack of focus aids/Af
I also think that future AF could have a big impact in the value of cinelenses and bodies like the aleza/F55 part of the reason Ive never jumped to any really big ticket cine kit as an owner op.
I predict that AF will sneak onto movie sets throught the back door, first used on Movis then for telephoto running shots then.. then..
S
Last edited:
Andrew Stone
Veteran
This is huge, if it works as billed and yes disruptive. Thanks for the info.
EDIT:
They will be at booth C9548
A few things that occurred to me about this unit is, you will probably be able to preprogram not only focus pulls but zoom pulls as well. Not sure if there is a ramping function that has been built into yet. I will certainly be asking about this.
I presume as well, you will be able to use the FStop wireless follow focus device/interface and your choice of digital motors with it.
Last edited:
That looks good from the tech stand point but I don't think these guys have a clue how(and why) focus pulling is done on set. I can't see how this will work in narrative or even docu style shooting. Focusing is a part of the story telling and the focus puller is an important part of the creative team. I can't see that aspect in the video. You need an IT pro to program this thing and I personally don't know any 1st ACs that are also IT pros. Remember what happened to oh so hyped Teradek encoder in the MP industry?
Andrew Stone
Veteran
The chess vid was troublesome in many spots, however, it demonstrates the tech. I can see practical applications where you aren't operating with razor thin focus situations, doing walk & talks in tight spaces.
To your last point, in the audio/mixing world they already have the protocols in place where you have automation happening on the mixing sliders (or pots) and you have an operator riding the sliders to tweak out the mix either in real time or after the fact.
This tool is not for the DSLR crowd. All kitted out this is going to be in the range of a Preston System. You have to look at it in general applicaiton and not shooting at F2.8 in close range doing 12 micro pulls in the space of 30 seconds.
I can see how anyone in the follow focus manufacturing world would be anxious about this product.
CharlesPapert
Director of Photography
Yes, well, that was the tech I was mentioning, I didn't realize they were releasing the info publicly before the show.
I had started to write a more generic set of thoughts about this issue last night but forgot to post. I haven't seen the Andra in person yet but have talked to their rep, so I don't know what the specific interface is for each function, so consider the below more of a generic discussion.
So there's a certain percentage of things to shoot where focus is basically a 1 or a 0, a binary equation, where the subject is either sharp (preferable) or not (bummer). For a human focus puller, it all comes down to skill at achieving consistent dead-on focus however they go about it: old school with marks and guesswork and intuition and maybe a Cinetape, or new school off a monitor. For an autofocus system, the goal is to achieve at least as well or better than a human, without error or hunting or being flummoxed by obstacles like a window in between the camera and subject.
Given this binary-type challenge, let's imagine a really tricky shot, such as a racecar driving at high speed right towards the camera at telephoto. An extraordinary focus puller may be able to manage the distant part of this using their bag of tricks, but even they will inevitably start to fail as the car gets closer and closer. The holy grail of this kind of shot would, for example, being able to maintain focus as the car closes in to where the frame starts to fill with the driver as seen through the windshield. Imagine how cool that shot would be--going from several car lengths wide to a closeup of the actor's eyes. The reason we haven't seen that without resorting to CGI is that it is all but impossible for a human being to manage that focus pull. However the technology to achieve this via a focus assist system exists, and is heading to market.
That's an extreme situation and again, no creative choices are to be made other than to keep the driver in focus at all times. When shooting narrative there are many decisions to make on where the focus should lie. We know that a subject's eyes need to be sharp for the subject to appear in focus, however there may be a situation where one has to make a decision between which eye that might be (with very shallow focus and the head slanted away from camera). A great focus puller knows how to work the depth of field so that the subject may be at 6 feet but he/she is actually setting focus at 5'7", because it will still hold on the subject but keep the foreground more in focus (or vice versa). And then there is the process of racking from one subject to another: while the two focus points may be an absolute, the timing and speed of the rack is a human judgement call that may alter from take to take along with the actor's rhythm. A fully automated AF system is going to have a tough time emulating the performance of a skilled focus puller when it comes to the organic choices.
Even with sophisticated tracking of a subject there are difficult situations that will come up, such as when they move behind someone else momentarily but you want the focus to pick them up as they emerge from the other side. A certain percentage of users will likely be quite satisfied with a system that works most of the time. In mission-critical situations, we are likely to see a fairly even tradeoff where an automated system is able to track a moving subject with more accuracy than an AC, but then it will fail at other predictive situations like the above by jumping to another subject, or "getting confused". Ultimately I think the ability to active the system as needed will be the MO for most situations where there is a skilled AC present.
I have long imagined a future where an AC will sit with a touchscreen tablet showing the live image, touching the screen at the next point of focus, then using a swipe to activate and set the speed of a rack. An instant rack could be triggered with a double tap. An integrated focus knob could be used to fine-tune the point of focus as an override, or program a move that can then be repeated by the software.
Finally, what of a documentary setting where the operator is pulling their own focus. The onus is generally on just getting the shot so the application of AF is quite desirable. There is still the question of how to "tell" the camera where the point of focus would be (imagine three people facing the camera at slightly different distances), perhaps this will be best served with eye trackinSg software that can place a focus box onto the desired subject. Presumably this will have to be turned on and off by the operator since one's eyes are usually darting around a frame. Not quite sure how one will do rack focuses with this interface. There's also the possibility of a mode where the operator pulls focus to the best of his/her ability and the AF system takes over by interpreting what they are "trying" to do and fine tuning the focus onto the subject.
I had started to write a more generic set of thoughts about this issue last night but forgot to post. I haven't seen the Andra in person yet but have talked to their rep, so I don't know what the specific interface is for each function, so consider the below more of a generic discussion.
So there's a certain percentage of things to shoot where focus is basically a 1 or a 0, a binary equation, where the subject is either sharp (preferable) or not (bummer). For a human focus puller, it all comes down to skill at achieving consistent dead-on focus however they go about it: old school with marks and guesswork and intuition and maybe a Cinetape, or new school off a monitor. For an autofocus system, the goal is to achieve at least as well or better than a human, without error or hunting or being flummoxed by obstacles like a window in between the camera and subject.
Given this binary-type challenge, let's imagine a really tricky shot, such as a racecar driving at high speed right towards the camera at telephoto. An extraordinary focus puller may be able to manage the distant part of this using their bag of tricks, but even they will inevitably start to fail as the car gets closer and closer. The holy grail of this kind of shot would, for example, being able to maintain focus as the car closes in to where the frame starts to fill with the driver as seen through the windshield. Imagine how cool that shot would be--going from several car lengths wide to a closeup of the actor's eyes. The reason we haven't seen that without resorting to CGI is that it is all but impossible for a human being to manage that focus pull. However the technology to achieve this via a focus assist system exists, and is heading to market.
That's an extreme situation and again, no creative choices are to be made other than to keep the driver in focus at all times. When shooting narrative there are many decisions to make on where the focus should lie. We know that a subject's eyes need to be sharp for the subject to appear in focus, however there may be a situation where one has to make a decision between which eye that might be (with very shallow focus and the head slanted away from camera). A great focus puller knows how to work the depth of field so that the subject may be at 6 feet but he/she is actually setting focus at 5'7", because it will still hold on the subject but keep the foreground more in focus (or vice versa). And then there is the process of racking from one subject to another: while the two focus points may be an absolute, the timing and speed of the rack is a human judgement call that may alter from take to take along with the actor's rhythm. A fully automated AF system is going to have a tough time emulating the performance of a skilled focus puller when it comes to the organic choices.
Even with sophisticated tracking of a subject there are difficult situations that will come up, such as when they move behind someone else momentarily but you want the focus to pick them up as they emerge from the other side. A certain percentage of users will likely be quite satisfied with a system that works most of the time. In mission-critical situations, we are likely to see a fairly even tradeoff where an automated system is able to track a moving subject with more accuracy than an AC, but then it will fail at other predictive situations like the above by jumping to another subject, or "getting confused". Ultimately I think the ability to active the system as needed will be the MO for most situations where there is a skilled AC present.
I have long imagined a future where an AC will sit with a touchscreen tablet showing the live image, touching the screen at the next point of focus, then using a swipe to activate and set the speed of a rack. An instant rack could be triggered with a double tap. An integrated focus knob could be used to fine-tune the point of focus as an override, or program a move that can then be repeated by the software.
Finally, what of a documentary setting where the operator is pulling their own focus. The onus is generally on just getting the shot so the application of AF is quite desirable. There is still the question of how to "tell" the camera where the point of focus would be (imagine three people facing the camera at slightly different distances), perhaps this will be best served with eye trackinSg software that can place a focus box onto the desired subject. Presumably this will have to be turned on and off by the operator since one's eyes are usually darting around a frame. Not quite sure how one will do rack focuses with this interface. There's also the possibility of a mode where the operator pulls focus to the best of his/her ability and the AF system takes over by interpreting what they are "trying" to do and fine tuning the focus onto the subject.
morgan_moore
Major Contributor
I dont know if you have used stills AF (i mean used it under commercial pressure - not for happy snaps) 'court and sport' - firstly one learns when it will work and wont.
For the car shot - well that is full AF and top canons and nikons can do it today.*
For the rest I think 'final five' (percent) is the way - you 'tell' the cam what you want in focus by making your best effort to focus on it - this then works as the talent gets crossed by an extra or there are three talent in shot.
To me the real focus issue can be just not being able to tell even with a static subject. (This often happens when the subject is 'static' (consistent distance) but moving (nodding dancing etc) creating motion blur in the frame) - I find the monitor (DP6) tears/strobes - maybe higher end monitors do not - I certainly feel more connected though a composite monitor which I feel has less latency.
I feel final five is required even with the best humans to do sharp 4k
Stills teles do have final 50% with switches restricting say infinity to 5m or 5m to macro - such restriction stops really bad hunting - that is a start but only a start
S
* a pelicle mirror motion camera could use top end canon and nikon AF at the cost of only one stop which is trivial in cams that will do clean 3200
For the car shot - well that is full AF and top canons and nikons can do it today.*
For the rest I think 'final five' (percent) is the way - you 'tell' the cam what you want in focus by making your best effort to focus on it - this then works as the talent gets crossed by an extra or there are three talent in shot.
To me the real focus issue can be just not being able to tell even with a static subject. (This often happens when the subject is 'static' (consistent distance) but moving (nodding dancing etc) creating motion blur in the frame) - I find the monitor (DP6) tears/strobes - maybe higher end monitors do not - I certainly feel more connected though a composite monitor which I feel has less latency.
I feel final five is required even with the best humans to do sharp 4k
Stills teles do have final 50% with switches restricting say infinity to 5m or 5m to macro - such restriction stops really bad hunting - that is a start but only a start
S
* a pelicle mirror motion camera could use top end canon and nikon AF at the cost of only one stop which is trivial in cams that will do clean 3200
Last edited:
CharlesPapert
Director of Photography
Sam, I don't use pro-level stills AF so I don't know what modes are available there.
I'm interested to hear that the car shot can be 100% achieved with current AF. Let's say we overcrank to 60 or 120 fps--are you saying that current AF systems can hold that in focus all the way through without wandering a bit?
I personally can't see critical focus on a 6" monitor regardless of performance. It's why I huddle in a tent with a 24" reference monitor, leaned in with my damn reading glasses on, day after day. I have noticed that my younger AC's like to pull from those smaller monitors (TV Logic is the most popular brand) but the more "seasoned" ones prefer 17" monitors, on the whole. They are all using Prestons, btw.
I'm interested to hear that the car shot can be 100% achieved with current AF. Let's say we overcrank to 60 or 120 fps--are you saying that current AF systems can hold that in focus all the way through without wandering a bit?
I personally can't see critical focus on a 6" monitor regardless of performance. It's why I huddle in a tent with a 24" reference monitor, leaned in with my damn reading glasses on, day after day. I have noticed that my younger AC's like to pull from those smaller monitors (TV Logic is the most popular brand) but the more "seasoned" ones prefer 17" monitors, on the whole. They are all using Prestons, btw.
morgan_moore
Major Contributor
A nikon D4 or canon will do a pretty good job on 100m sprinters at the olympics with a 1/2 body as the last frame. F1 car into a full helmet shot at 200MPH, well maybe not 
They use prediction, ie distance at time A was X, distance at time B was Y, distance at time C was Z therefore one can extrapolate the acceratation (using differential equations) that given a consistent acceleration, which is more complex and accurate than calculation based on consistent speed) the position at C+1 will be Z+x where x is the result of the calculus
So it works well with cars, runners and the like and less well with brownian objects (like actors?)
If you have a 24 monitor.. lucky you it has to help
S
They use prediction, ie distance at time A was X, distance at time B was Y, distance at time C was Z therefore one can extrapolate the acceratation (using differential equations) that given a consistent acceleration, which is more complex and accurate than calculation based on consistent speed) the position at C+1 will be Z+x where x is the result of the calculus
So it works well with cars, runners and the like and less well with brownian objects (like actors?)
If you have a 24 monitor.. lucky you it has to help
S
Last edited:
CharlesPapert
Director of Photography
If you have a 24 monitor.. lucky you it has to help
Ain't no luck Sam, just ROI choices! Monitors have been a good one--My 24" FSI and the two Marshall 17"s for video village paid themselves off in about a year.
morgan_moore
Major Contributor
starcentral
Veteran
There's a 1st AC & I bring on projects from time to time and shooting in 4k on the F55 he pulls focus from a 7" monitor with 1280x800 resolution and uses no gimmick features like peaking from either the camera or the monitor.
It was on one particular project where he operated the camera on a jib arm, pulled focus, and did so using Nikon lenses (reverse focus direction) that I quickly realized how highly skilled some guys are with it, and how poorly skilled at it I am.
It was on one particular project where he operated the camera on a jib arm, pulled focus, and did so using Nikon lenses (reverse focus direction) that I quickly realized how highly skilled some guys are with it, and how poorly skilled at it I am.
Last edited: