Panasonic 4K GH4 possibly coming next Spring...

I think understand how downsampling 4K 4:2:0 can result in 2K 4:4:4. However, I don't understand how an 8-bit image can become a 10-bit image regardless of the resolutions involved.

It's my understanding that bit depth is directly proportional to dynamic range, ie. an 8-bit image has approx. 8 stops of dynamic range. Which is why the BMC has 13 stops DR since it's capable of 12-bit raw images. So if a 4K 8-bit image is downsampled to 2K then wouldn't the dynamic range remain unaffected?
 
When you downsample 4k to 2k, each group of 4 pixels from the 4k image becomes 1 pixel in the 2k image. When you combine 4 8-bit quantities, you end up with a 10-bit quantity.
 
gkantziper said:
I'd be curious to hear from others on the forum about this idea: The GH4 is potentially 3 stops faster @ 0 db gain than an AF100 which is rated 400 ISO @ 0 db gain.

I'm GUESSING this based on this information:

An AF100 can be gained up to 18 db / 3200 ISO and the GH4 can be gained up to 25,600. So, if 25,600 is 18 db on a GH-4???

Panasonic states, "GH4 enables shooting at maximum ISO 25,600. Sensitivity, gradation performance, resolution and color reproduction are dramatically improved to achieve even higher picture quality."

Correct me if I'm wrong, but it follows the GH4 is 3200 ISO at 0 db gain. That'd be nice but it isn't likely true!
Click to expand...

There's no reason to think that the GH4 is limited to 18dB of gain though. It might be capable of much more than that. The HMC40, for example, could go to 24dB, and the AC90 I think goes to 27 or 30dB. So you can't infer 0dB ratings off of the potential maximum, unless we knew for a fact that the specifications stated a maximum of 18dB of gain.

I would expect that the GH4 could have a higher base sensitivity, if for no other reason than that it's a few years newer sensor technology. But we can't put a finger on what the real base rating is based off this speculation, we'll need to see the actual specs.
 
noirist said:
When you downsample 4k to 2k, each group of 4 pixels from the 4k image becomes 1 pixel in the 2k image. When you combine 4 8-bit quantities, you end up with a 10-bit quantity.
Click to expand...

I don't believe that is true. When you interpolate from 4k 8bit to 2k, you end up with 2k 8bit, you can't add any new information, because it was lost when you originally sampled at 8bit, all you would be doing is shoving an 8bit sample into a 10bit package.
 
atarijedi said:
I don't believe that is true. When you interpolate from 4k 8bit to 2k, you end up with 2k 8bit, you can't add any new information, because it was lost when you originally sampled at 8bit, all you would be doing is shoving an 8bit sample into a 10bit package.
Click to expand...

As a simplified illustration: imagine having a square of 4 pixels, each of which can only be black or white (1-bit depth). If you average them, you have 5 possible outcomes: black, 1/4 gray, 2/4 gray, 3/4 gray, white.
 
IanTheM said:
So that your computer doesn't have to spend all night converting footage, and you can get to work editing right away. Not to mention the storage.
Click to expand...
Premiere can't handle the 4K down-scaling natively without seperately converting? And what do you mean "the storage"? 1080p at 200 mbit/s is twice the storage space as 4K at 100 mbit/s, so you'd actually have half the storage space to deal with.


joe 1008 said:
For the ease of postproduction. The 200 mbit/s codec stores each single frame while the 4K100 codec is la long GOP format that only stores the difference between frames. Decoding such 4K footage will require a hell of a computer and so does downscaling the footage to 1080p 4:4:4. Also have in mind that the 200 mbit/s can store up to 96 fps while the 4K codec only goes up to 30 fps. So 96 fps in 1080p is roughly the same data as 24 fps in 4K. Having in mind that the long GOP in 1080p, which is also able to store up to 96 fps, also goes up to 100 mbit/s the numbers start to fit, don't they?
Click to expand...
I wonder how good of a computer. People often converted AVCHD and h.264 footage but I never have using Premiere and a $2,300 Windows laptop I got a year and a half ago which runs those files smoothly. I'm all for not converting footage so if a better computer is all that is needed that's an option.
 
atarijedi said:
I don't believe that is true. When you interpolate from 4k 8bit to 2k, you end up with 2k 8bit, you can't add any new information, because it was lost when you originally sampled at 8bit, all you would be doing is shoving an 8bit sample into a 10bit package.
Click to expand...
Each of the 4 super-pixels can have a value 0-255 in each of the three colors. Let's pick a color, say Green. If all 4 super-pixels have 0 Green, then the down-sampled pixel will have a 0 value as well. If there are three 0s and one 1, then the down-sampled pixel will have value 1. If there are two 0s and two 1s, the down-sampled pixel will have value 2. If there is one 0 and three 1s, the down-sampled pixel will have value 3. If there are four 1s, the down-sampled pixel will have value 4. Now do the math for all possible super-pixel values and you will see that the down-sampled pixel can have values between 4x0 minimum and 4x255 maximum. The range of down-sampled pixel values is from 0 to 4x255, which requires 10 bits to represent. QED.
 
Resolve 10 Lite is free and handles 4K now (Free version does)

I believe it would do quite a job compressing the 4K files to ProRes 1080 or 2K for you guys, you'd just need to get your computers ready to handle that.

More memory on your GPU = Better

I've recently (finally) had to start updating everything for a 4K workflow, and that meant going with a GeForce Titan to start with.
 
noirist said:
Each of the 4 super-pixels can have a value 0-255 in each of the three colors. Let's pick a color, say Green. If all 4 super-pixels have 0 Green, then the down-sampled pixel will have a 0 value as well. If there are three 0s and one 1, then the down-sampled pixel will have value 1. If there are two 0s and two 1s, the down-sampled pixel will have value 2. If there is one 0 and three 1s, the down-sampled pixel will have value 3. If there are four 1s, the down-sampled pixel will have value 4. Now do the math for all possible super-pixel values and you will see that the down-sampled pixel can have values between 4x0 minimum and 4x255 maximum. The range of down-sampled pixel values is from 0 to 4x255, which requires 10 bits to represent. QED.
Click to expand...
Thanks for breaking it down. It makes sense to me. I also found this nice article on chroma subsampling. The important thing is the pictures in it helped me visualize how 4K/UHD 4:2:0 could translate into 2K/FHD 4:4:4: http://www.pcmag.com/encyclopedia/term/57460/chroma-subsampling
 
If you down convert 4K to 2K to get the better color space, can you still crop in post, or have you now lost your 4K resolution such that you'd be cropping in on a 1080 image?
 
Eric Coughlin said:
If you down convert 4K to 2K to get the better color space, can you still crop in post, or have you now lost your 4K resolution such that you'd be cropping in on a 1080 image?
Click to expand...

You now have a 1080p image, so any cropping will be on a 1080p image just like cropping from a current 1080p camera. Except you'll have a higher quality 1080p image. Akin to something that was shot with a 3 chip camera but with more luminance as if there were a fourth chip (a second green for instance).
 
So basically if you convert from 4K to 2K to get the better color space and then you have shots you want to crop in on, you should use the original 4K 4:2:0 files on those shots as better resolution will likely be more important (particularly on heavy cropping) than better color space.
 
Or just use the original files the whole way through and just in the end, squeeze it to 1080p. I doubt they are that much harder on your computer than avchd. Maybe slightly.
 
Eric Coughlin said:
So basically if you convert from 4K to 2K to get the better color space and then you have shots you want to crop in on, you should use the original 4K 4:2:0 files on those shots as better resolution will likely be more important (particularly on heavy cropping) than better color space.
Click to expand...

Technically, you are getting better color resolution. The color space stays the same.
 
[h=1]Which 4K camera for the masses? GH4 vs Blackmagic Production Camera[/h]
Conclusion
The internet tends to polarise our arguments and unique, specific uses of a camera get buried under the headlines. The message here is clear – both cameras offer a great deal to filmmakers, whether it’s the flexibility of a raw workflow and great software like Resolve 10, global shutter’s ability to capture extremely fast movement as the eye sees it, or the GH4′s amazing all-round feature-set and convenience factor.
Price wise the Blackmagic is expected to be launched at around 3700 euros and $3995. That’s considerably more than the mooted launch price of under $1999 for the GH4. Add to the Blackmagic your SSD drives to store all that raw and ProRes, plus external battery and articulated EVF or monitor and you have a much more expensive piece of kit.
The GH4 has a lower point to entry but to get 10bit 4:2:2 and ProRes from an uncompressed 4K feed you will need the add-on box. This will put the camera in the same price bracket as the Blackmagic Production Camera and ideally require the same outlay on SSDs and monitors. At least you get two tasty XLR inputs thrown into the bargain and a full sized HDMI port.
Overall the GH4 is the most appealing for me, but your needs may differ and GH4 users without an external recorder may look on jealously towards the 10bit colour of the Blackmagic camera come the shipping date.
Click to expand...
http://www.eoshd.com/content/12057/4k-camera-masses-gh4-vs-blackmagic-production-camera
 
noirist said:
Each of the 4 super-pixels can have a value 0-255 in each of the three colors. Let's pick a color, say Green. If all 4 super-pixels have 0 Green, then the down-sampled pixel will have a 0 value as well. If there are three 0s and one 1, then the down-sampled pixel will have value 1. If there are two 0s and two 1s, the down-sampled pixel will have value 2. If there is one 0 and three 1s, the down-sampled pixel will have value 3. If there are four 1s, the down-sampled pixel will have value 4. Now do the math for all possible super-pixel values and you will see that the down-sampled pixel can have values between 4x0 minimum and 4x255 maximum. The range of down-sampled pixel values is from 0 to 4x255, which requires 10 bits to represent. QED.
Click to expand...


For the GH4, there aren't 4 "super pixels" with 3 complete full resolution R,G,B channels to start with. 4:2:0 footage is Y'CbCr subsampled, the CbCr channels already have 1/2 resolution width and height . So essentially you have 4 Y' "super pixels" which can be expressed at 10bit greyscale 0-1023 as per the illustration when scaled down to HD, and 1 Cb "super pixel", 1 Cr "super pixel", which become 8bit color difference channels 0-255 .
 
You must log in or register to reply here.
Back
Top