jcs

On a 12-Core MacPro (2.93GHz), SSD drives, getting 1.62 FPS converting DNG to 50Mbps H.264 (16% CPU utilization) in AE CS6. AE also leaks memory along the way (didn't run out of RAM (24GB) for 641 frames, but got close). The test shot of the equipment on my desk looked very nice (dropped frames; boring, not posting it ;))- sharp, detailed, clean (NR with ACR), even slightly higher res compared to my FS700 + Speed Booster. The 128GB Komputerbay 1000x card maxed at around 55MB/s, so can only do 1920x720 max. This is almost half of the advertised 90MB/s min write speed shown on Amazon.com (sending it back). Trying a 32GB card next (Komputerbay rates at 100MB/s). I don't see using this for anything other than really short takes given the current workflow, so 32GB is fine for now. Given my knowledge of GPU shaders and the DCT used for JPEG, it should be possible to convert directly from RAW to MJPEG faster than real-time (I haven't written a GPU debayer routine yet, though the basic concept is simple enough. Not clear how much work to get something close to ACR quality (low-artifact sharpness and NR might be challenging)). Sony has implied that the FS700 will get XAVC in the upcoming firmware update next month. The hardware upgrade so far sounds to be around $300 (might be required to get the firmware update). If the FS700 gets 10-bit 422 H.264 (XAVC), not sure if the extra work for the 5D3 RAW workflow will provide much improved results for most shots and content targeted for the internet. The FS700 has builtin ND's and pro audio, another major bonus for fast workflows compared to the 5D3. Then there's the slomo too... a lot more cost, but worth it (4K has yet to be seen, but could be very useful too (with Convergent Design hardware)). Still dig the 5D3 stealth factor and small/light handling in general (along with an IS lens).

  Fixed.   Regarding cranky folks on the internet- understand that they are in pain for one reason or another (mind, body, or spirit). Don't take their comments personally- they tend to be cranky to everyone. Being kind to them anyway can help them out of their pain. Minimally, at least you'll feel better.

Here's a simple test I did last year with real-time film grain using the GPU, include a noise sweep from left to right, then a noise pattern from 0 to 100%. For the first part, it's subtle and makes it through the H.264 encoders. For the 2nd part, most of the high frequency components are lost (through DCT quantization).   http://www.dvxuser.com/V6/showthread.php?278876-Film-Grain-Real-time-GPU-PPro-CS5-5   There's a .zip file there with the GPU accelerated PPro effect and instructions how to use it. By tweaking this GPU effect (including changing the blend level and blend type), every time I compare tweaked results to the Filmconvert (demo), I don't have a need to purchase it. I like the product/looks, but it's not GPU accelerated in PPro. Since I have very limited time to shoot and edit, I set a constraint to only use real-time tools and effects. I've looked through the PPro/AE SDK and have not seen any plugin example code which can take as input an OpenGL texture and send a modified texture back (required for full GPU acceleration). Not clear if Adobe allows this for 3rd party plugins. Have been writing GPU shaders for iPhone/iPad and would love to write some for PPro (not interested in AE- not a real-time tool).

If the 5D3 provided full HD resolution (1920x1080 vs. 1620x910), a higher bitrate codec (422 LGOP 50Mbps or better (same as C300)), and pro-audio (XLR) input with decent preamps, it would be game over for new products for a while at that price point. That's why Canon hasn't done it. The closest we can get to that right now is an FS700 (or FS100) with a Speed Booster (if full frame look is a goal. Internal codec good enough for many applications and external recording via SDI/HDMI works well). Black Magic cameras without more efficient codecs, better low-light performance, and better ergonomics, have limited use (+ way too much data for most applications). Modular cameras systems are the next step (sensor separated from recorder). Decoupling the sensor from the computer and lens provides a lot of flexibility.

A decent implementation of IPB will always look better than I-Only if IPB is allowed up to I-Only bitrates. For fast motion (even moderate motion), the information that lower bitrate IPB has to throw away to hit the target bitrate generally can't be seen when played in real-time (except in extreme cases or bad implementations). I tested IPB vs. I-Only when the 5D3 first was released. IPB was clearly more detailed and had less macroblocking (I posted frame grabs as well): http://www.dvxuser.com/V6/showthread.php?278746-ALL-I-vs-IPB-IPB-Wins This could have changed with the new firmware. When I want the best quality I skip the 5D3 and use the FS700 with Speed Booster and Canon lenses (the FS700 isn't perfect but produces a real ~1920x1080 image vs. the 5D3's ~1620x910 image).

A 3D LUT is all that's need to get any skintones you want (color curves and grading will work too). The problem with bright blue is interesting- shared with the FS700. Fortunately pretty rare, however hoping that Sony will fix in future (firmware or HW fix).

  1.3 crop comment: it seems that the 1DC and 5D3 aren't that different. The 1DC's full frame image looks very similar to the 5D3 (soft), whereas the ~1.3 crop and S35 mode looks pretty good.

The reason it's hard to see a difference between 422 and 420 is exactly the reason they toss the color information (1/2 the vertical color information to be exact). The only visible difference between 422 100+Mbps Sony MPEG2 (Nanoflash) and 420 24Mbps AVCHD on the FS700 is less compression artifacts and more high-frequency detail (typically noise though). The advantage for the 5D3 would be the same: less compression artifacts (and more noise as the H.264 codec naturally does a good job reducing noise (high-frequency detail)).

Thought maybe it was my setup (Nanoflash) but your footage also shows a green => magenta shift from HDMI to H.264.

It works with the Nanoflash: 24p. Live preview on a Samsung D8000 looked really good (shooting a model up close). Didn't see a way to enable camera display with HDMI out at same time. Didn't appear to send audio out HDMI (perhaps is turned down or off- need to explore). Zero increase in resolution vs. internal: A/B test in Premiere. Color and brightness shift internal vs. external (perhaps Nanoflash issue (HDMI Color: was on Full vs. Legal)). Nanoflash at max bitrate (180Mbps) had major macroblocking vs internal IPB (1080p24). Camera stationary on desk. Have other tasks to take care of, then can explore more.

Hardware update required for 4K: http://www.dvxuser.com/V6/showthread.php?305725-4K-will-require-hardware-update Not clear if only for Sony's recorder or for 4K in general.

In the video he said "in the internal XAVC(!) codec)"!!? That would be an awesome, welcome upgrade, especially if 10-bit!

In addition to the F55, the F65 also has a global shutter. Regarding 4K for the Production camera: how many photosites? Only the F65 comes close to actual 4K resolution. All the other 4K cameras are really 3.xK. To get the film look it's very important to have high resolution without any aliasing or digital artifacts.   Personally I'd rather have a camera that used high-bitrate H.264/H.265 + 422/444 10-12bit vs. fat+inefficient ProRes HQ (effectively ancient MJPEG)). These are cool cameras for sure, but they won't be replacing reliable pro gear such as the F5/F55 (could see some RED sales going this way though). Feels like a niche product and not really an upgrade for DSLRs given the SSD workflow, form factor, and massive files generated (including ProRes HQ vs. any long GOP codec: XAVC or similar would be a welcome addition).

Nice- ProRes HQ also supported. If they can make it reliable and ship on time, very cool.

Don't have time to read through all the pages- not clear if this has been mentioned. This system is not gyro stabilized in the traditional sense with cameras (e.g. Kenyon labs style using high-speed spinning masses). This type of stabilization doesn't even need a gyro- all that's needed is an accelerometer and filtered magnetometer. From these two sample vectors, an orthogonal basis can be formed whereby a rotation matrix can be created. Simple pots or optical encoders can be used on the gymbols and from there the math can be done to tell the servos how to move the camera. Since mems gryos are so cheap, it's natural to use them to improve orientation determination quality (accelerometers have motion bias and magnetometers are noisy). Notwithstanding any patents, this type of system could be built fairly inexpensively- the highest cost elements would be the motors/servos and the case/frame to mount the camera. The electronics would be low cost- same price as RC equipment.   $100- same concept: http://diydrones.com/profiles/blogs/705844:BlogPost:2185 Note comment regarding gyro drift: this is where the accelerometer (gravity) and magnetometer (compass) can be use to prevent drift (added cost, but still cheap).   [video]http://www.youtube.com/watch?feature=player_embedded&v=YJXv-Qm4ZUc[/video]   $3500 for something more advanced (as mentioned by others: Zenmuse): http://aerialtechnology.com/shop/zenmuse-3-axis-gimbal-sony-nex-7/ With more competition, the price will come down.

First watched Hertzfeldt's work (Rejected) in San Diego at Spike & Mike's Sick & Twisted Animation Festival in La Jolla. My sister "rejected" Rejected as appropriate for her kids to watch. I showed it to them them anyway- they loved it. I still crack up remembering what she said, "Seriously John. 'My anus is bleeding'. That's one of the worst things I have seen in my life. I want my 10 minutes back". Yeah, she watched the whole thing, lol.   Regarding Vimeo and PPV for indie content- great idea: needs to be on living room devices and iOS/mobile, the same as Netflix, and a means to quickly surface the good content and hide the bad stuff. The problem with most content (including studio content) is it sucks (as others have pointed out). Good content, if it can be surfaced, will always sell.

Saw The Croods in RealD 3D, AMC ETX in Burbank: Sony 4K and Dolby Atmos 3D sound. The screen was as big as the last IMAX screen I viewed (huge). We sat towards the back on the upper balcony- image was tack sharp. 3D visuals were perfect and 3D audio was very cool. When this quality is available in the home, folks will buy it. Current 3D and HD TVs cannot provide this kind of experience (major issues with 3D at home are low resolution or ghosting). Conversely, saw the Life of Pi in 3D at the Dome Arclight in Hollywood with shutterglasses- clunky, heavy, uncomfortable 3D glasses, ghosting, not a good 3D experience. The movie was great however I can see why folks don't like 3D if that's all they have seen (low quality 3D systems).   I have worked with 3D for a long time- my company created some fo the first 3D games for PCs (including kicking off hardware accelerated 3D by porting Quake to 3D in 1998 for H3D). Have seen 3D come and go; it's not going anywhere this time around: it will slowly get better in the home. No pundit or hater proclamations will make it go away. RealD 3D rocks- check it out before writing off 3D. 4K will be a small market for a while but makes sense for large screens.   Clearly, 3D is not a proven failure: http://www.youtube.com/watch?v=Lzrmx6HhtyQ   If you don't get it, see: http://www.youtube.com/watch?v=MuOvqeABHvQ   If you don't get that, you're not doing enough acid*       *ascorbic acid. What did you think I meant?

What's your budget?   Who's going to fly it?   A GoPro Hero 3 Black (2.7K mode, scaled down to 1920x1080) will probably be fine with a much cheaper copter.

Clean 422 with an external recorder will certainly help with reducing macroblocking and can provide better quality for post-sharpening. Without some kind of improvement in actual sampling resolution, my FS700+Speed Booster makes the 5D3 only appealing for close work and/or stealth shooting (in which case the external recorder kind of hurts stealth recording)   If it's not possible to keep the camera display active when using an external recorder (such as the Nanoflash), then stealth recording won't be possible: would have to use an external display as well.

Kind of odd, showcasing a sensor optimized for video recording static stars. If resulting products can produce true "Full HD" (true 1000+ lines H & V sampling resolution without aliasing), will be pretty cool. Since optimized for video only, could use the same photosite arrangement as C100/C300 for true Full HD (4K sensor). The 5D3, 1DX, and 1DC do around 800 lines for full frame HD vs. the required 1000+ lines for Full HD (the 1DC can do this in 4K mode sampled down to HD; S35 crop mode appears to be somewhere in between- a chart test could show us for sure).

I believe sensor noise is constant. You can push down the amplitude of the noise and increase the apparent signal to noise ratio by using curves/profiles to push down the lower intensity output as with an S-curve (will increase contrast and push the blacks down, along with noise). Brighter source pixels hide noise better (higher signal relative to constant noise), thus raising the output of brighter pixels won't increase perceived noise as much.   When using a highly compressed output as with 24Mbps H.264, it's best to get the look as close as possible in-camera, with limited post-grading. This is because the output is 8-bit and the DCT quantizer throws away a lot of information during compression. If you need for example the blacks to be higher, you'll also notice more noise during recording. However, you can remove the noise in post with Neat Video and still preserve decent edge detail, with limited banding. If you shoot with a profile that hides noise during recording, and you try to raise the blacks in post, you might end up with large blotchy areas and banding. Using an external recorder with 422 and a higher bitrate can also help.   For the FS700, it's probably wise to create a few different profiles, one for each common shooting condition. You've got 6 slots, so that should be plenty.

Nice to see post-sharpening (PS) being embraced as a viable tool when comparing camera performance, Andrew. As long as no artifacts are surfaced, PS can do a much better job than in-camera methods. By turning off/down in camera sharpening there could also be an advantage in effective noise reduction and edge preservation due to the way DCT quantization works. While the 5D3 can indeed produce sufficiently sharp images with PS, if the shots aren't perfectly focused with sharp lenses, especially for medium to wide shots, PS can't bring the shot up to a sufficiently sharp level and looks soft compared to other cameras (noise grain can help, but only for brief clips). The 5D2 is a 1.4-1.5K camera (aliasing improves perceived resolution without post-sharpening) and the 5D3 is a 1.6K camera at best with PS: not much room for error and not sufficient for wide shots with far-away objects. From both a technical and cost standpoint, there's no reason in 2013 for camera manufacturers to not be able to deliver full, true, 1000+ h & v sampling resolution 1920x1080 video cameras with little or no aliasing (oversampled at least 2x in hardware along with a sufficient OLPF). 10-16-bit HDR delivered with H.264/H.265 should be the next product differentiator.

That's the "ARRI Alexa" setting  ;)

I would have considered the C100 (even the C300) if they had full frame sensors. I prefer Canon's sensor design and slightly higher horizontal resolution, however the FS700 + Speed Booster is a super versatile camera and allows the use of existing lenses (without having to purchase more wides). Thus, the FS700 + SB makes a nice 5D3 upgrade. It also works very well for ENG/run&gun as the autofocus and stabilization of the kit lens works very well. The new Sony SELP-18200 lens has servo zoom and will work with the FS700's zoom rocker. If it's fast enough, could work well for "The Office" kinds of shots. I'm not aware of any autofocus/zoom lenses like this for Canon's large sensor cameras.

Based on the way the patent claims are written, unless the patents are invalidated in court, RED can comfortably own the compressed RAW (6:1 or greater) market. It appears as long as the sensor data is de-Bayered to RGB/YCrCb, no infringement. 12-16+ bit H.264 (or H.265) would be pretty nice and without the massive file sizes: I would prefer that for my FS700 vs. the planned "4K" RAW support.   BMCC is OK as it's uncompressed RAW.