I suspect it's a true 5.7K Bayer sensor, like the GH5 is a true 5.2K Bayer sensor. The 5.7K size is to help improve the detail and color of the finished de-Bayer'd 4K image. ( same way the GH5 does, same way Sony does with their A6500 camera )
EVA1: Panasonic AU-EVA1! Announcement and latest updates
- Thread starter soarprod
- Start date
- Status
I suspect it's a true 5.7K Bayer sensor, like the GH5 is a true 5.2K Bayer sensor. The 5.7K size is to help improve the detail and color of the finished de-Bayer'd 4K image. ( same way the GH5 does, same way Sony does with their A6500 camera )
filmguy123
Veteran
+1 and I've been one of the most vocal about internal 10bit 4K 60p. Don't get me wrong, I really wish the EVA-1 could do it, but Thomas is spot on: there's way more to a camera than specs. Color rendition, menus, workflow, reliability, ergonomics, size/weight. The way you work with a camera and the total package is far more important than one technical spec on a spreadsheet. EVA-1 looks very promising in all of these categories, and that's extremely important.
Well, we'll see soon enough and the images can speak for themselves. I will say the increase in detail (not SHARPNESS, but detail) from the GH4 to the GH5 which downsampled was huge. Night and day difference. Because of the debayer process it seems to really help to have the image downsampled. Not an engineer so can't speak to your point on if that would be better to do in post or in camera, but all I can say is the final image on the GH5 which employed the technique was a big improvement, even in FHD.
Mitch Gross
Veteran
That 5.7K resolution is the photosites on the sensor. Simple as that, and no interpolating at all.
Briefly, here's how it works. If we were shooting monochrome, it would all be easy. A photosite senses brightness and the resulting image would match one photosite to one pixel of the finished image, so a 4096x2160 sensor would yied 100% resolution on a 4096x2160 finished image. But we like color, so the math isn't so simple.
A photosite can only sense brightness, so where do we get color info? In the old days we used to use three sensors that were aligned using a prism block so that one did red, one blue, and one green. Work out the brightness values for each and that's what's sent out as the final pxel signal, so three photosites generated one pixel. But now we use Bayer-pattern sensors, where a single sensor is used with a color filter array spread (CFA) overtop. The CFA has a row that's blue/green/blue/green, followed by a row that's green/red/green/red, and then repeating. So while a 4K sensor for 4K image has resolution data for every pixel, it does not have color data for every pixel. This color data is created for each pixel location by averaging the information from the photosites surrounding that location. The math is extremely accurate, but in the end there is less data that is being used. The estimate (and it must be an estimate as there are both various ways to do the math and the accuracy of the result depends on the nature of the incoming image) is that the resulting image resolution is about 70-75% of the native photosite resolution.
So here's where the math gets fun. Why did Panasonic go with a 5.7K sensor? 70% of 5.7 is 4. By using a 5.7K sensor, one is able to achieve a fully-resolving 4K finished image even when taking into account the losses inherent in a Bayer-pattern sensor.
Matt Gottshalk
Resident Preditor
G
Guest
Guest
This was what I was thinking it was. That is why I was wanting clarification. A lot of people are assuming this camera will have a 5.7K full color image that they can crop in post or re-sample in a 4K workflow.
In a previous comment you said 5.7K raw output. Sounded like you were saying 5.7K final image.
But it is in-fact a 5.7K monochrome sensor at the heart, but a 4K final output in color. Thank you for the clarification.
FWIW, GH4 tested out to about 1,500-1,550 lines off 8.8 MPX (center crop) but DVX 200 was right around 2160 off 13.2 MPX (a much smaller center crop). Hypothetically speaking the whole debayer/demosaic process could be repeated, if there's enough processing power. On other occasions, the Bayer pattern is different from the one Mitch described, as in Fuji cameras and F65.
Barry_Green
Moderator
Well -- you could. Just as a Red One was a "4K" camera, in that it was 4096 x 2160 and people used it to make 4K images, you could just as easily do "5.7k" on the EVA1 if you record raw. Same as on a BlackMagic Ursa, it uses a UHD sensor to make UHD color images, so you could certainly get a full color 5.7K image out of the EVA1.
Now, that 5.7K image won't be quite as detailed as, say, an 8K sensor downsampling to 5.7K, no. Just like the Ursa's image isn't as sharp as a 5K or 6K Red that downsamples to UHD.
The Ursa has a 3840 x 2160 sensor. The EVA1 has (about) a 5700 x 3200 sensor. So yes the EVA1 will be extraordinarly sharp hyper-sharp 4K, much sharper than an Ursa or a "true" 4K sensor could make. And yes, you can make bigger images with it. The EVA1 raw should make images that are comparably sharp to a 6K Red.
All sensors are monochrome at heart. All sensors produce monochrome images that (with the addition of the CFA and some wizardry in demosaicking) turn into color images, at somewhat lower resolution than the native sensor pixel res. That's why the DVX200 has a 5K sensor, so it can make extremely detailed 4K images.
Again, just compare it against the competition -- the EVA1's sensor is 5.7k. That is going to be much sharper than the Ursa's 3.8K UHD sensor, or the Ursa 4.6's 4.6K sensor, or the Red Raven's 4.5K sensor, or the Scarlet-W's 5K sensor, and it will be extremely close to the 6K off a Red in terms of resolved detail.. Whether we know anything else about the EVA1 or not, I think we can confidently say that resolvable detail is not going to be an issue, this is going to be a hella sharp camera.
Last edited:
G
Guest
Guest
Thank you for the thoughtful response.
But the main concern I had here was not about sharpness, but rather a workflow one. Handing off files with odd pixel counts can be a hassle. And I was worried there were going to be soft 5.7K files that were a waste of space. So, I am glad to hear that the EVA is a 4K camera. I wouldn't be opposed to 5.7K, I just have no need for it, and I am glad it is 4K.
With Mitch's "5.7K output" comment and several other people, it was starting to sound like this was an 8K sensor with a true 5.7K output. But I knew that was ridiculous, so I was prodding for clarification on what was meant by 5.7K. Mitch did just answer that question a few posts ago. It is a 5.7K sensor, with a 4K color file output.
Again, based on several comments on these boards, and from people in town, there was the idea that this was going to be an 8K sensor to 5.7K output camera. But Mitch kindly just clarified that it is a 4K camera. Even the Panasonic literature says "downsampled from 5.7K". That doesn't seem like the correct way to describe the process of debayering.
Yes, resolution was simpler discussion in the 3CCD days. haha. I own an MX camera, and very aware of the softness in 4K.
That is exactly what I was hoping for. It is great being able to shoot with true resolution. Soft images can be a waste of data rates, and over sharpened images are ugly. In an ideal world, the resolution is real.
But the main concern I had here was not about sharpness, but rather a workflow one. Handing off files with odd pixel counts can be a hassle. And I was worried there were going to be soft 5.7K files that were a waste of space. So, I am glad to hear that the EVA is a 4K camera. I wouldn't be opposed to 5.7K, I just have no need for it, and I am glad it is 4K.
Last edited:
Mitch Gross
Veteran
One of the options for RAW output will be 5.7K. RAW is ALWATS the native photosite resolution from the camera. That's the definition of RAW; it means pre-deBayered. From that 5.7K RAW one can do many things. One could generate a 5.7K full color image that will have the same color and resolution accuracy as a 4K image that is derived from a 4K sensor, meaning photosite-to-pixel resolution. From that one could choose to pan & scan if so wished. One could also choose to supersample the sensor native resolution, taking the 5.7K RAW data and rendering a 4K RGB image. For that matter one could also choose any other lower finished resolution if they wanted.
Please do not attribute to me any false equivalencies or other words I did not write. Internally the EVA1 will utilize the 5.7K sensor to achieve high quality 4K, UHD, 2K, 1080p or 720p images. EVA1 will also utilize the 5.7K sensor to output those resolutions as video over SDI & HDMI. In a later firmware update, the EVA1 will be able to output RAW data from the sensor at various selectable resolutions, including 5.7K, 4K, and 2K. When selecting those resolutions, the RAW data output will be the native photosite resolution directly off the camera's sensor.
Please let me know if any of this is not clear.
G
Guest
Guest
Very clear. Thank you.
Thank you for your replies.
If you felt I was attributing false statements to what you said, I am sorry.
I was quoting you, and describing my perception of what you were saying. Not to accuse, but to clarify.
Jumping back and forth between camera, lighting, sound, and post, terms get fluid.
Raw footage in editing can mean 'uncut' and unorganized. But raw, is also the term for minimally process video and photo files.
So, while raw is not an acronym, people often spell it as RAW, because it is helpful. Still, it is weird that something incorrect can be adopted to help clarify a term.
The term "4K output" is similar to me in that way. When I am sending a signal or data to people, and I tell them it is 4K output, it had better make a 3840 × 2160
color file for playback.
Sony chose to say 20megapixel sensor for 4K for the F65. This is clearer to understand since "K" is more often used as a format than a senor's native resolution. Whereas Megapixel is ubiquitously used to describe sensor resolution, and not commonly used to describe a video format.
The other term "downsample". It is usually referred to taking a 4K 4:4:4 file, and turning it into an HD 4:2:0 file, for example. But if "down sampling" is used to describe the debayer process, then it is confusing whether it is taking a larger image and resampling to a lower resolution, or whether it is just debayering.
Last edited:
Thomas Smet
Veteran
Hey I also wanted 4k 60p at 10bit but we didn't get it this time. I also don't view it as a total loss. Many of us have shot stunning material in 8bit for years. 8bit does not equal bad. 10bit doesn't make video magically look better. In most cases it just has extra overhead to provide better sampling in really complex situations like shooting a subtle gradient in a sky. Not all skies look bad in 8 bit either. 8bit vs 10bit has been debated for a very long time. There are those that assume it should magically look better right away but it doesn't. It's just extra overhead. Just like 24bit audio is extra overhead compared to 16bit audio and 14 stops of DR is overhead compared to 12 stops. ProRes is overhead compared to h264. All these things can look good in most conditions. It is those rare conditions where things like 10bit help.
If you shoot carefully the 8bit should be perfectly fine just as it has been for Sony A7S users, Canon c100 users and any other 8bit camera out there. To this day the GH4 and now GH5 are the only DSLRs in the world capable of some kind of 10bit output and yet DSLRs are used for some very high end work. I'm not saying 8bit is as good and anybody who it is is not viewing what the true potential of 10bit really is and why it is used.
If the shot isn't too complex for 8bit then visually there should be little to no difference in a 32bit float grading environment. The 32bit float interpolates any in between samples needed which is why people have been so confused thinking 10bit provided no advantage over 8bit. It just meant they didn't shoot anything complex enough to break 8bit. Where 32bit float can't help is if the shot is already damaged and has banding. Banding are gaps in the samples and you can never go back and fill those in. If 8bit has no gaps however you can push the life out of that shot with 32bit and it will look great.
V-log does complicate the 8bit space and you do have to be a lot more careful with v-log in 8bit but I would not call it unusable. Depends on the shot really. Just throwing out a hypothetical number here but if 10bit v-log can safely get 98% of the shots then 8bit v-log can get 80% of the shots. You are at a slight disadvantage with 8bit v-log in some situations but not all situations. Fortunately Panasonic managed to get external output at 4k 60p 10bit so if that requirement is critical for you there are options. I know not optimal if you don't want to lug around a recorder but it is an option to go along with all the great benefits the EVA will provide over other cameras. The oversampling and dual ISO are kind of worth the price on their own for anybody who wants to do 4k delivery.
Im interested in seeing how the 4K 60P holds up in 8bit. My experience says when shooting 8 bit, better get that image as close to the final in camera. That means accurate exposure and white balance. Limiting the gradients and complexity of the shot. These are all achievable of course and accurate exposure and white balance should always be the aim.
EDIT: What Thomas said pretty much.
EDIT: What Thomas said pretty much.
Barry_Green
Moderator
In terms of megapixels, the EVA 1 should be somewhere around 18 megapixels.
Regarding video frame sizes, traditionally "video" has been described in terms of vertical pixels (480i, 720p, 1080i, etc) whereas digital Cinema formats have usually embraced the horizontal pixel count (2k, 3k, 4k). Generally those "k are in terms of 1,024 (so 3k = 3,072 and 4k = 4,096, etc) although there is no hard and fast rule; UHD is generally used interchangeably with 4k to describe the 3,840 x 2,160 frame size.
I have never heard of "downsample" being used to describe debayering. As far as I know that term is or should be used specifically for describing the process of resizing an image to a smaller resolution.
Mitch Gross
Veteran
James, I don't want to pick on you in any way, but there are some assumptions and/or inaccuracies in your post. Yes, many technical terms get used improperly, and we should be diligent in striving for clarity and accuracy. I delve in here to use this as an opportunity to educate all, as there's plenty of misunderstanding on these subjects by many people. To that end, a few comments.
In both still photo and motion picture work, RAW (all capitalized to distinguish it from other uses) means footage that is shot with a Bayer-pattern or similar sensor that has not yet been de-Bayered. It can be compressed, it can be noise-reduced, it by definition has gone through an A/D conversion (analog to digital). It simply means not-yet-deBayered.
There are two variants of 4K. DCI is 4096x2160, and is the theatrical format (DCI = Digital Cinema Initiative). UHD is 3840x2160, which maintains the 16x9 ratio of standard HD. UHD is also known as QHD (Canon) and QFHD (Sony), as technically UHD (Ultra High Definition) is supposed to be an open term that can be used for even higher resolutions, such as 8K and beyond (Lordy help us). BTW, these distinctions exist in HD as well, where DCI 2K means 2048x1080 and FHD is 1920x1080 (Full High Definition, for those playing along at home).
The F65 uses a different layout of photosites to gather data. Imagine rotating a sensor 45 degrees, so that the square photosites are "on end" like diamonds. The apply a Bayer-pattern mask to this. That's the design behind the F65 sensor. It actually works quite well, but it is a very different bit of math to get those tilted photosites to represent square pixels in the end. But the point is that there are a LOT more photosites then there are pixels, so the data is available and it works.
Then there is "downsample." You may notice that I did not use the term. I wrote SuperSample. A downsample is a way to get a smaller image from a larger starting image. It also doesn't have to be an image, as the term exists in audio and frankly just about any other form of cumulative data. To downsample one stays within the same media form, but goes to a smaller container, so information is averaged and thrown away. Downsampling has nothing to do with deBayering. SuperSampling is different. In SuperSampling, one goes to a smaller container, but also better utilizes the available data to yield the highest quality result in the smaller format. An example would be starting with a 4K 10-bit 4:2:2 image and ending with a 2K 12-bit 4:4:4 image. While the resolution has decreased, the bit depth and the color sampling are improved by utilizing the available brightness and color data in the original image. That's the distinction. As far as the 5.7K RAW to 4K video conversion goes in the EVA1, there are two processes at play -- deBayering as well as SuperSampling. While there is interaction the two are different.
For contrast, an example of downsampling is what happens in the Sony A7-series cameras when switching between 4K UHD and HD. In both resolutions the signal is limited to 8-bit. Compare that to a feature in the Odyssey7Q+, which can accept that camera's UHD 8-bit data and SuperSample it to become an HD 10-bit. This is a noticable improvement in image quality.
Mitch Gross
Veteran
Dontcha all love it when Barry & I are both up way too late, posting away here?
G
Guest
Guest
If i said you said it, i am sorry. I was referencing the panasonic page for the EVA
http://business.panasonic.com/AU-EVA1.html
"Downsample to 4K" is used in the panasonic site:
Last edited:
Quick update here. Thanks for the feedback everyone. Here's what is going to happen.
Rather than place EVA1 into Varicams we are going to change things to reflect how Panasonic refer to their cameras. Varicam heading will go and will be replaced with Panasonic Cinema Cameras under which we will have prefix for Varicam35, VaricamLT and EVA1.
Cheers
Noel
graywolf
Well-known member
Maybe I am misunderstanding something, or maybe some of you guys are.
The EVA1 is quoted as weighting 2.63 lbs, that is closer to 3 than the 2 people keep saying. The stripped LT is supposedly 2.4 lbs heavier. That comes out to 5.3 lbs in my math.
EVA1 under $8000, stripped LT $14000. Now if you add those $10000 worth of accessories to both you get $24000 14 lb LT. or an $18000 12 lb EVA1. Lenses are going to add $10000 to $50000 to those prices.
The actual savings seem to be minor, 2 lbs and $6000 off, maybe a $50000 total for the Varicam LT.
So what am I missing here, in an apples to apples comparison?
Justinec101
Active member
Yes if you add $60,000 of stuff to the Eva the savings ratio decreases.
- Status