AG DOK - Arbeitsgemeinschaft Dokumentarfilm e.V.
Schweizer Straße 6
D-60594 Frankfurt am Main
Telefon: +49 69 623 700
AG DOK-BVK CAMERA TEST 2019
CAMERAS WITH FULL FRAME SENSORS
Kristina Kozlova (model)
Christopher Rowe BVK (DoP, planing, article)
Felix Trolldenier AG DOK (DIT, colorist, analysis, article, planing)
Rasmus Sievers AG DOK (operator, planing, article)
Erik Wittbusch AG DOK (DIT, operator, planing)
Anne Misselwitz AG DOK (operator)
Gerd Breiter AG DOK (operator)
Steph Ketelhut AG DOK (operator)
Colin Elves (operator)
Robert Löved von Slashcam (operator)
Jonas Spähinghaus (assistant)
Eberhard Spreng (set fotos, article)
Shoichi Sano (assistant)
Natalie Pusch AG DOK (assistant)
Tim Haber (assistant)
In cooperation with
Equipment kindly provided by
The current test is dedicated to the new cameras with full-frame sensors, inflated called "large format" by manufacturers. Aside the predictable optical properties of a larger recording format, we examined how the new sensors behave in regard to skin tone reproduction, contrast range and low light. We also looked at the image quality at higher frame rates.
The cameras: Arri Alexa LF, RED Monstro VV, Canon C700 FF, Sony Venice, Kinefinity Mavo LF, for comparison, a Sony a7s II with external recording and as a representative of the Super35 class a Ursa Mini Pro 4.6K and an Amira.
In addition, we will again demonstrate the advantages of the ETTR exposure method and show what lies behind the "LF look".
We shot the following setups:
Setup 1: Studio-Porträt mit Tageslichtscheinwerfer
Setup 2: Verfügbares Licht bei Nacht
Setup 3: Studio-Porträt bei Kunstlicht
Setup 4: Zeitlupe
Links to the videos can be found in the relevant sections of the article. Here are all the videos as a playlist:
Bitrates of the codecs used in comparison:
The following matrix gives a rough impression of the performance requirements of the various codecs (recording formats) for a computer.
Programs are increasingly using the GPU to decode, so that lower demands can be made on the CPU, or computation time becomes free for e.g. more layers in grading programs.
The sensor sizes in comparison:
The sensor sizes in comparison:
The Mavo LF, Venice and A7SII have a sensor matching exactly the 35mm format. The sensor of the Alexa LF is slightly larger. The one of the C700 FF and especially the Monstro are wider than full frame, but less high. While C700 FF and Monstro would be pillarboxed for the standard aspect ratio 1.85:1 with spherical lenses, the other cameras would need letterboxing. The usable sensor surfaces in the test most closely correspond to the VistaVision format: 36x18.3mm:
A larger sensor sees a larger image with the same focal length. In order to keep the image section constant (with the same camera position or perspective), a longer focal length is necessary. Multiplication by the format factor (crop factor) results in the new focal length. It is calculated e.g. from the relationship between the diagonals of two formats. 35mm small picture / full format has side lengths of 36x24mm and thus according to Pythagoras a diagonal of 43.27mm.
An aspect ratio of 1: 1.85 results in the following format factors:
Differences in aspect ratio of 1.85: 1 are therefore insignificant in practice for these sensors. In order to set up the same shot, in our studio motif, e.g. the difference of the distance of the sensor plane and model between the Alexa LF and the Monstro was only 8cm.
SETUP 1 STUDIO PORTRAIT WITH A DAYLIGHT SOURCE
Relative f-stop values
Measurement at 1/50s, ISO 800: Führung 16, Aufhellung 5.6
Als Führung wurde eine Arri M18 HMI als Booklight verwendet, d.h. mit der Lampe eine Styro angeleuchtet und das reflektierte Licht mit Voll-WD Frostrahmen (#216) gestreut. Mit einem Styro auf der Schattenseite wurde im Gesicht ein Kontrast von 1:8 (3 Blenden) erzeugt. Die Spitze und das Licht auf dem Hintergrund waren ebenfalls HMIs. Die Lampe im Bild war eine stark gedimmte 300W Arri Kunstlicht Stufenlinse.
Spektrum gemessen mit einem Jeti Sepcbos 1211 gemessen auf Weißkarte: 5550K
The cameras were - in contrast to previous tests - not adjusted to the actual color temperature but set to 5600K preset. In steps of 1 stop, we drew i.d.R. Exposure levels from -6 to +6.
The brightest exposure at which the top lights in her face have not yet clipped, we call the ETTR exposure (see explanation in test 2016).
VENICE [LOW NATIVE ISO]
VENICE [HIGH NATIVE ISO]
MAVO LF [LOW NATIVE ISO]
MAVO LF [HIGH NATIVE ISO]
URSA MINI PRO
ETTR vs Standard exposure
Coincidental same exposures 1
Coincidental same exposures 2
Mavo LF vs Monstro + Noise Reduction
SETUP 2 AVAILABLE LOW LIGHT
Unfortunately, C700 FF is not included here due to an exposure error. Low-light comparisons extracted from Setup 1 (see above) and Setup 3 include the C700 FF.
SETUP 3 LIMITED INCANDESCENT LIGHT
Measured spot values
The guide was a dimmed 1kW RifaLight (Softbox), with a measured color temperature of 2400K. It was not possible to position the color chart to give it the same light as the model. It is additionally lit by an undimmed 300W Fresnel lens. The light on the background produces an unfiltered HMI, the light strings on the left in the picture are commercially available "warm white" LEDs with a clear green cast, on the left in the picture is a tungsten light tube.
Measurement: guide 125 Lux (EV 8 2/3 = T2.8 at 1/50s & 800 ISO)
Spectrum measured on white card: 2370K
White balance: 3200K Preset
Three typical lighting situations should be simulated to test how good the color rendering of the cameras is at low color temperature, and what effect underexposure has on image quality. Particularly interesting was the question, which advantages the cameras offer with dual ISO in low-light situations.
The 3 exposure levels:
+/- 0: a typical night / indoor illumination where there is enough light available for a "normal" (no ETTR) exposure
-2: an "available light" situation eg. in a typical bar. The available light is not enough for a "normal" exposure
-4: an "available light" situation eg. under street lamps. Turning under these conditions is justified only by the content
During exposure, the amount of light was adjusted according to the native ISO specification of the camera manufacturer. Thus, at the level +/- 0 the sensors of the 800-ISO cameras were exposed with 1/50 - T2.8, the Venice (Native ISO 500) and Mavo (treated by us on the basis of pre-tests deviating from the manufacturer's specification such as ISO 500) with 1/30 - T2.8 and the Sony A7SII (native ISO 1600) with 1/100 - T2.8.
At level -2, the settings for Alexa LF, Monstro and C700FF were:
ISO 3.200 - 1/50 - T2.8 - ND 0.6.
The Mavo camera was set to 5,000 ISO, so the "Dual native ISO" was automatically activated, but in accordance with the already mentioned evaluation of the sensitivity of the camera (2/3 aperture weaker than specified by the manufacturer) as 3.200 exposed, ie with the same settings as the other cameras
On the Sony Venice, the "Dual Native ISO" was set to 2,500, and the exposure changed accordingly to 1/40.
At level -4, the ND filter was changed, ND 1.2 instead of 0.6, all other settings remained the same.
In addition, we recorded the -2 with the same exposure for all cameras.
For the Sony A7S, increasing the ISO value has a positive effect on noise, so we only tested the higher ISO values for the same amount of light for this camera, that is, ISO 25,000 at the -2 level and ISO 100,000 at the level -4.
The ISO setting on Alexa LF, Monstro and Venice has no meaning for RAW codec recording and only applies to metadata. For the C700 FF and Mavo LF, the ISO knob acts as a pre-gain before saving the RAW data.
AG DOK-BVK CAMERA TEST SETUP 3 LIMITED INCANDESCENT 0 F-STOPS
AG DOK-BVK CAMERA TEST SETUP 3 LIMITED INCANDESCENT -2 F-STOPS
AG DOK-BVK CAMERA TEST SETUP 3 LIMITED INCANDESCENT -4 F-STOPS
AG DOK-BVK CAMERA TEST SETUP 3 LOW LIGHT 1 (selbe Belichtung)
AG DOK-BVK CAMERA TEST SETUP 3 LOW LIGHT 2 (selbe Belichtung)
The best results of the two underexposure levels produce the highest ISO (gain). In RAW, both with "baked-in" ISO and metadata, Gain spreads the signal and assigns more data to it. These help in the Grading usable signal to separate from the noise. The A7SII adds internal noise reduction with increased ISO. For all clips in Post, a white balance was performed using the gray patches on the color chart with the RAW settings and the curves in Scratch. The Master Luma Curve adjusts the contrast of the clips and sets the face to a maximum of 80% in the waveform.
The high proportion of infrared in the spectrum of light sources probably contaminates the names, visible in the color changes of their blue dress. The monstro seems the least affected by the infrared radiation. At least z.T. the IR filtering could help the Schneider NDs.
SETUP 4 HIGH SPEED
Increasing frame rate is often accompanied by a reduction in resolution. Particularly evident is the increasingly chromatic aliasing of the water droplets in the photographs of the monstro. Exception is like in the 2016er test the Arri, which records also with 150fps 4.5K RAW without quality loss. However, Mavo LF and Monstro achieve higher frame rates.
Some manufacturers succeeded in successfully combining properties such as "3D" or "lifelike" images with "LF" cameras. However, a larger format, even with an equivalent focal length, does not possess such optical properties.
The geometrical description of the Thin Lens explains in first order the relations between format size, format factor, focal length and aperture, here for example the S35 format with a 33mm objective and full frame format with 50mm objective.
We assume here that the goal is to reproduce the identical shot with a different format, i. E. the camera position (perspective) remains the same.
The diagrams show the simple relationships between format sizes, focal length and aperture: If the format is increased, the focal length must increased as well to obtain the angle of view. This proportionality is expressed by the crop factor - the ratio of height, width or diagonal of two formats. In the example here: 1.5. A full-frame camera with a 50mm lens sees the same shot as a Super35 camera with a 33mm lens.
Not only the equivalent focal length but also the equivalent f-number is calculated using the format factor. The goal is to keep the diameter of the blur circles constant in the images of both formats.
The following formula and diagram on Wikipedia describes the diameter C of the blur circle depending only on distances of object (focus) S1, S2 - camera and diameter of the absolute aperture A.
The relative aperture or f-number is in turn proportional to the focal length - determined by the format factor.
So, with a 35mm sensor, you'll gain a little more than an f-stop of blurriness (DOF) over the S35 format. In other words - with the (longer) equivalent focal length on the full frame camera, a smaller aperture suffices for the same depth of field. In this example a 2.1 instead of a 1.4.
Two optical characteristics of full frame format can be noted:
1. The larger format provides about 1 stop shallower depth of field at the same equivalent focal length and aperture.
2. Maximum apertures of available optics in the market for s35mm and full size are at the same level, i.e. with full frame sensors, it is easier to produce shallow depth of field.
Furthermore, "LF look" nonspecific is associated with "less distortion", other "rendering of space", more separation in depth and more resolution.
Distortions: There are two types of distortion. Optical and perspective distortion.
Optical distortion: barrel and pincushion distortion of lenses can be corrected and depends on the position of the aperture in the optics. By recording a raster, a correction file can be created. For the integration of VFX a daily process and does not represent an in "LF" inherent, or at smaller formats not reproducible, property.
Perspective depends only on the position of the camera.
Perspective distortion is especially noticable in short-focal length shots, like a nose, might appear unnaturally large. This is acutally only caused by small distance of camera-to-object, regardless of format or focal length. The camera renderes the image correct, but the eye would see the same shot differently, as the brain would deskew the image, for a more natural picture.
The other extreme is the seemingly increasing compression of the space behind objects in the foreground, which is not caused by longer focal lengths, but also only by the distance of the camera to the object. A crop of a wide-angle shot is identical to a telephoto shot taken from the same position.
Depth separation: This is depth of field (DoF) and it is reduced, as shown above, with increased format size, at equivalent focal length, constant aperture and object distance.
More spacial resolution by more pixels of a larger sensor may be advantageous, albeit only for oversampled aliased reduced images. Unlike larger film formats, digital cameras with larger sensors do not necessarily offer more resolution (see RED Helium for example) or lower noise.
With a constant distance to the object in focus, the same f-number, equivalent focal length (same image section) and circle of confusion, it is only the reduction in depth of focus that causes the larger format.
Mehr zu dem Thema von Steve Yedlin ASC.
The cameras were shot one by one with the new 40mm Sigma FF High Speed Cine Primes. Differences of the sensor sizes we compensated with little change of the camera position. Exposed was usually with T4 or T5.6, only at the extreme overexposure and underexposure levels the lens was opened to T1.5 or closed until T11. The exposure time was usually between 1/50s and 1/100s, with strong overexposure or underexposure to 1/25s or 1/400s. If available, internal ND filters were used, for Alexa LF, Red Monstro, Mavo LF and Sony A7S we used Schneider Rhodium FSND filters in the strengths 0.6, 1.2 and 1.8. Both the internal filters and the Schneider NDs had turned out to be color-neutral in a preliminary test.
Grading was done in Assimilate Scratch primarily with the master luma curve, RAW settings and the manufacturer's LUT, with the aim of preserving the character of the color science of each camera.
The Lanczos filter was used for scaling. Material was recorded or debayred with sharpness off and without any noise reduction unless stated otherwise.
Arri Alexa LF
The Arri still offers (despite presumably unchanged sensor design since 2010) the largest dynamic range. Also, their colors remain relatively constant over its range. In the warm light of setup 3, however, it drifted towards yellow, which however could be corrected in the RAW settings.
The original Alexa with s35 sensor with its roughly 3K sensor is the standard for larger productions. It offers sufficient resolution for the big screen and generally best picture quality. The Alexa LF with the same and larger sensor with more pixels now meets the formal 4K requirement of Netflix with 4.5K.
The weakness of the sensor is low light, even with -2 stops underexposure, the picture begins to show noise - whether the pleasant or "cinematic" is a matter of taste. At least at -4 aperture, the picture without noise reduction in the post is not recommended. For night photography in the presence of light, dual ISO cameras are available, e.g. the Sony Venice or the Mavo LF. The Panasonic Varicam, which stood out in the 2016 test by very good low-light properties at 2,500 ISO, is currently offered only with S35 sensor.
The usability is as usual from Arri very simple and clear, who knows the Alexa is immediately familiar with the menu. The RAW recording works smoothly. The quality of manufacturing is Arri typical high and robust.
The Alexa LF was the heaviest camera in the test. It sits very well on the shoulder and gives through the well-distributed weight handheld camera images peace and stability. But you do not want to operate off the shoulder for a longer period of time without a support system.
Negatively, we noticed the lack of an internal ND filter. Instead, Arri offers a system with which ND filters can be mounted in strengths 0.3 - 2.4 behind the lens. We could not test this system, but it is clear that a filter change behind the lens e.g. in a dusty environment is not advisable at all and even under optimal conditions is much more time consuming than to replace filters in front of the lens in the Compendium.
Only after our test came the Alexa LF Mini on the market, it is not only small and lighter, but has as usual of the Alexa Mini and Amira internal ND filters.
The LF mount is too smooth. The combination of PL and LF with two identical lock systems can lead to errors. Working with LF Mount lenses is not a problem, but is a weak point in combination with PL lenses. Several times we had the LF mount in hand, although we have solved only the PL mount.
Video: Amira vs Alexa LF
Apart from the Mavo LF, it is the smallest and lightest camera tested, ideal for handheld cameras, gimbals, small spaces and discreet filming.
The combination of R3D and simultanious ProRes recording provides a solid workflow for editing. To a certain degree the 8K resolution allows in-post underexposed settings to still be saved via noise reduction.
Red improved handling with the new body. Although in terms of usability, the Monstro gives a mixed impression.
It is easy to use. The menu, however, reacts slowly. Changing the WB, ISO, frame rate and shutter speed is easy with the LCD display - once you get used to the menus or the jog wheel. Also, the variety of slow motion options (and the speed changing it) makes them a very versatile camera.
The Monstro is flexible in the choice of optics: The option to prune the sensor, as well as the choice of mounts (EF, Nikon, PL).
On the other hand, however, it has many disadvantages that limits its applications, especially for documentary work. Shooting in low light may include an OLPF switch for optimal skin tones, including a 15-minute recalibration of the sensor. You do not want to do this in a dusty environment. Likewise, the need to toggle sensor calibration files when the shutter angle is changed (eg, for Slomo), is prone to errors and time consuming.
The lack of internal NDs requires a matte box and the time needed to change NDs. The lack of decent internal sound recording capabilities (no built-in XLRs, poor internal preamps, no external audio control) and fan position and noise indicate that this is a full-crew camera: operator, assistant, mixer.
Unlike typical random variable pattern noise (VPN), in the underexposures of setup 1, horizontal lines appear through the image and are more reminiscent of digital RF interference.
CANON C700 FF
The C700 FF is slightly lighter and slimmer than the Alexa LF, but for longer free camera work, especially with the Codex recorder attached, still too massive.
After the color science of the C300 II disappointed us, we were looking forward to a new sensor. Unnecessary gamma / gamut / matrix options, already found in the C300 II, again required a pre-test. After analysis, we opted for
Curve: clog2 (largest dynamic range)
Matrix: off (eos original for more saturation and compatibility with C300 I)
sharpness: 0 (or -10)
The camera shows with increasing underexposure more and more magenta in skin tones, but then drifts extremely green and loses much saturation in the other channels. Whereas she disappointed in her low-light performance. Setup 3 shows Sensorsmear, which also the C300 II suffers from.
On the assistant side there is another menu display. Increasingly, the manufacturers are here to align themselves and are based on ARRI. Submenus remind of the C300, though, are scattered on different menu structures, which is confusing.
In addition to the Venice, she was the fastest to record directly after triggering.
The camera offers dual recording, the position of the CFast slots, however, need getting used to.
Less practical for smaller projects is the uncompressed RAW with 5Gbit/s, but the C700 FF offers a variety of XF-AVC bit rates up to 910 Mbit/s.
Usablity, weight and capabilities give lets the Venice appear as a cheaper and handy alternative to the Alexa LF. In particular, their above-average lowlight performance in the test field makes her interesting. Sony-typical Magenta skin was corrected with the new s709-LUT, which produces an overall warmer image, maybe a bit too warm. Its colors remain relatively constant over its dynamic range and it shows even in underexposure good color separation. However, a variable pattern noise clouded the overall good impression.
The control panel is on the right side. The menus have become clearer by aligning with the ARRI style. Unfortunately, it reboots with some changes to the project settings without warning. The Sony's own X-OCN XT codec is more efficient compared to ARRIRAW and no visible quality disadvantage in our clips.
Disadvantages: The camera does not perform internal oversampling on smaller pixel dimensions. So it only records higher frame rates in 1:1 pixel crop modes. Highspeed is only available up to 60fps. A firmware v4 will offer up to 120fps, probably in the summer of 2019.
Due to its low price, this camera is particularly interesting. Good colors and, above all, the best low-light performance among the large cameras in the test make it stand out. Her dynamic range is only average though. A wide selection of codecs allows for flexible tailor-made workflows and her low weight and form factor allow for a wide variety of ápplications. The numerous mount options as well as the mountable electronic ND filters are ahead of the competition. All of this has to be weighed by each individual with the still existing problem of reliability. Extensive testing of the intended workflow, SSDs and monitoring is mandatory.
Overall, the Mavo LF looked like in beta stage. Our copy already showed hardware errors. Sometimes we could not turn on the camera because it did not boot up with the HDMI cable plugged in. This was allegedly related to the already defective video-out module of our test device. Though we had problems with crashes only in the pre-test.
In combination with the Sigma primes, the Mavo LF produced extreme aliasing, rendering its images practically unusable. Until an adequate OLPF is offered, the use of softer optics, filters and ProRes recording could reduce the artifacts. However, the OLPF is simply screwed "just" in front of the sensor and would be relatively easy to replace.
We did not test the e-ND in the mount adapter (PL or EF). However, we know from previous cameras to a good implementation in the handling: A pressure on one of the speed dial buttons, then you can adjust the wheel in ND in 0.03 steps (virtually infinitely). The ND takes a minimum of 2 stops.
Also untested was the Kineback-W with SDI and XLR.
In direct comparison with the images of other cameras showed at 800 ISO underexposure of the image, based on the WFM signals of the gray chart. As a consequence, we set the camera to native ISO 800, but exposed it as 500 ISO, 2/3 f-stop brighter. However, due to the good low light performance, this measure must be questioned.
Of the two SSDs available to us, only one SSD cDNG could record 1:3 without interruption. The official 120 GB kinemag we used is still the first generation and is not officially suitable for the Mavo LF. Kinefinety offers only 500GB and 1TB models for the camera.
As attractive as standard commercial SSDs appear as a recording medium, it requires research and own tests at the desired data rates.
The many different crop modes and frame rate combinations according to (old Terra6k) manual could not be adhered to in the test. Again and again stopped recordings and displaying "FLUSH". Only all ProRes 10bit variants worked reliably. Possibly the older Kinemag was the problem.
Another special feature is the high frame rates of the Mavo with high pixel dimension, though sensor crops. At s35 it achieves at wide screen aspect ratios and 4K 100fps, and in M3/4 3k @ 150fps and in HD 200fps.
No currently available SSD allows cDNG to record 1: 3 in 50fps.
A total of eight basic functions such as ISO, Shutter, e-ND, LUT are located on the four directions of the click-wheels on the handle. After some familiarization, however, this allows very quick access to everything essential. However, the zoom function on the handle, whose range is selectable, only works smoothly with ProRes. With cDNG the magnifying glass was hardly usable.
Overall the handling was pleasant for a solo operator and reminiscent of RED, but seems to be more user-friendly, contrary to voices online.
Sharpness and exposure assessment with on-board monitor, however, is still a problem:
- no usable waveform
- zebra-function illogically implemented and inaccurate
- hard to getting used to peaking, not adjustable
Good, clear menu. Very good are the options for presets, with which you can quickly switch by 1-click between e.g. full frame 25p cDNG and M4/3 96fps ProRes switch.
Practically are the built-in LUTs (adjustable) with exposure compensation in 0.3 aperture steps, but unfortunately described in the menu unfavorable.
The A7SII benefits from its internal noise reduction at higher ISO settings, but creates a soft and image without texture in combination with the weak internal 100Mbps XAVC-S codec (8bit 4: 2: 0). At least the codec can be bypassed by external recording, but the artifacts of the noise reduction remain in the HDMI signal.
The external recording goes hand in hand with the known disadvantages:
HDMI connection an menu settings are prone to errors. In addition, nothing can be attached to such cameras without a cage.
In extreme underexposures, without raising the ISO, random jumps in color appeared. Maybe an artifact of noise reduction that switches below a signal strength. Such exposures are useless anyway.
In Setup 3, the new s709 LUT was used for the evaluation. It suprisingly produces better skin tones also with the Alpha.
URSA MINI PRO 4.6K [SUPER35 SENSOR]
The Ursa and Eva1 are the camera in the entry-level production segment, which we currently recommend. We shot with the Ursa only setup 1 for comparison.
Concerning dynamic range, she joins the weaker cameras in the test field. Starting at -3 stops, or ETTR -7 even the "Pro" version of the Ursa Mini still shows some pattern noise. It is also noticeable that with increasing underexposure it strongly raises the blue, and somewhat the green, channel. In the videos it is corrected by curves to see the noise. Their resolution is as expected lower than the 6 and 8K cameras.
More about the Ursa Mini Pro, refer to Test 2017.2:
from the left: Jonas Spähinghaus, Gerd Breiter, Erik Wittbusch, Anne Misselwitz, Felix Trolldenier, Christopher Rowe, Rasmus Sievers, Shoichi Sano
The AG DOK test team is always only volunteering and is independent of the manufacturers. However, the model received a salary from the associations.