blackrat Posted May 23, 2013 Share Posted May 23, 2013 Interesting...the GH2 is very close to the 5D raw in these tests and it doesn't need all this post processing and storage power. I understand the gravity of ML and their work, especially if you're an owner of a 5D or have a lot of Canon glass. However, for us on the fence between GH3 and 5D3 what is the allure of Raw? Considering it is so intensive of a workflow process it would only be used for special shots, and the majority of footage would be h.264, wouldn't the no brainer be to upgrade to a gh3? GH2 doesn't have the DR or color of a hacked 5D3. It's more than just the resolution that gets improved with the 5D3 hack. Quote Link to comment Share on other sites More sharing options...
blackrat Posted May 23, 2013 Share Posted May 23, 2013 Again I can't imagine any appeal to having to work with a completely different mount and camera system. If you have two Canons already just stick with Canon. They are the leader for a good reason, and glass is a major Canon strength. BTW, the internal ND's on the C100 just beat out Formatt and Schneider 4x5.65's in a shootout: http://www.ryanewalters.com/Blog/blog.php?id=6201317295579746489 Saves you even more dough (not to mention the also world-class built-in IR filter that also seals the sensor area from dust), and no flattening polarization effect or color bias from a Vari-ND. Really given we have two excellent indie-affordable options in the C100/Ninja2 (best lowlight) and FS700/Speedbooster (overcrank) it's heroic but puzzling of you guys to stick with DSLRs. The RAW hack is definitely interesting to make use of the 5D3 as a super-B cam instead of just stills. But even so kitting it out is more expensive than the C100 option if you are only interested in motion, and not as great an image as charts reveal. What if you also do a lot of stills? Then you need the 5D3 anyway which makes the costs even far more extreme. 5D3 gets it all in a small package. Great for hiking out to film cool spots/nature. Don't need to always lug two systems, etc. C100 still doesn't grade quite as well. I mean each have their uses though of course. Quote Link to comment Share on other sites More sharing options...
JHines Posted May 23, 2013 Share Posted May 23, 2013 You don't want to try to remove vertical banding with standard spatial NR! As you say that destroys all regular detail to only just remove some fixed pattern banding. You need to use special banding-tuned NR. Where would I find special banding-tuned NR? Quote Link to comment Share on other sites More sharing options...
JHines Posted May 23, 2013 Share Posted May 23, 2013 This chart should help with knowing what ISO's to use to reduce the presence of the fixed vertical noise patterns. Looks like 100, 200, 400, 800, 1600 and 3200 are the cleanest. Curious to try shooting a daylight wide landscape with ISO 100 and 800 to compare the vertical patterns. Seems like once you get above 800 it's less apparent as all the noise kinda blends together. I was shooting at 160 for my first raw test video and I'm pretty sure that ISO made the vertical banding worse. Quote Link to comment Share on other sites More sharing options...
peederj Posted May 23, 2013 Share Posted May 23, 2013 This chart should help with knowing what ISO's to use to reduce the presence of the fixed vertical noise patterns. Looks like 100, 200, 400, 800, 1600 and 3200 are the cleanest. Curious to try shooting a daylight wide landscape with ISO 100 and 800 to compare the vertical patterns. Seems like once you get above 800 it's less apparent as all the noise kinda blends together. I was shooting at 160 for my first raw test video and I'm pretty sure that ISO made the vertical banding worse. Interesting results though I always want to know how they were made precisely. It's not at all clear to me that the noise floor is the source of the vertical lines problem on an even gray field. Though NR will likely clear that artifact away if the NR is trained to it. One thing that does display, if accurate, is the standard ISO's 100, 200, 400, 800, 1600 and 3200 are the only ones you should be using on the 5D3 RAW. Above 3200 is just a digital push I understand and that looks like it there...it's like Digital Zoom, you would be better off doing it in post than burning it into the capture. The standard ISOs up to 3200 are gain applied to the sensor itself and so they do have benefit. In fact I understand there is quite a bit of read noise on the 5D3 that limits low-ISO DR more than it should and much more than the D800 does. The one hassle would be not being able to see what you're shooting very well in Live View...it might be a nice feature request to ask for from ML if the camera can't do it already (brighten the Live View picture without raising the recorded ISO past 3200). I'm glad we're getting more reproducible data to discuss, though the reproducibility part remains elusive. Keep it coming, and in exhaustive detail por favor. Quote Link to comment Share on other sites More sharing options...
nigelbb Posted May 23, 2013 Share Posted May 23, 2013 Which hack was used for the Gh2? thanks It's irrelevant. GH2 resolution is not improved by any hack. Unlike Magic Lantern which is reverse engineered new firmware providing a lot of new functionality the GH2 hack just alters internal registers that control how the image is compressed & recorded. Ernesto Mantaras 1 Quote Link to comment Share on other sites More sharing options...
Richard Collins Posted May 23, 2013 Share Posted May 23, 2013 Again I can't imagine any appeal to having to work with a completely different mount and camera system. If you have two Canons already just stick with Canon. They are the leader for a good reason, and glass is a major Canon strength. BTW, the internal ND's on the C100 just beat out Formatt and Schneider 4x5.65's in a shootout: http://www.ryanewalters.com/Blog/blog.php?id=6201317295579746489 Saves you even more dough (not to mention the also world-class built-in IR filter that also seals the sensor area from dust), and no flattening polarization effect or color bias from a Vari-ND. Really given we have two excellent indie-affordable options in the C100/Ninja2 (best lowlight) and FS700/Speedbooster (overcrank) it's heroic but puzzling of you guys to stick with DSLRs. The RAW hack is definitely interesting to make use of the 5D3 as a super-B cam instead of just stills. But even so kitting it out is more expensive than the C100 option if you are only interested in motion, and not as great an image as charts reveal. It's really because the 7D just doesn't cut well with the C100 and Ninja. The image stinks when compared. Quote Link to comment Share on other sites More sharing options...
peederj Posted May 23, 2013 Share Posted May 23, 2013 It's really because the 7D just doesn't cut well with the C100 and Ninja. The image stinks when compared. Everything else under $10K is going to. The 5D3 HDMI or RAW won't be as clean and high-res as we see but at least it can be wrenched into place to crosscut with it, especially when using Canon glass of the same marque (i.e. L's or non-L's). The RAW leaves lots of latitude to align with. The Panasonic color science is dismal, I mean, gross greenish yuck, you won't want that. I mean, look how lovely that C300 frame is, that's pro gear and the rest are hobby toys, including the clunky BMCC. Consider avoiding 2000+ quid of disappointing "bargains" a 2000+ quid discount on something that's actually going to work. Richard Collins 1 Quote Link to comment Share on other sites More sharing options...
Richard Collins Posted May 23, 2013 Share Posted May 23, 2013 Everything else under $10K is going to. The 5D3 HDMI or RAW won't be as clean and high-res as we see but at least it can be wrenched into place to crosscut with it, especially when using Canon glass of the same marque (i.e. L's or non-L's). The RAW leaves lots of latitude to align with. The Panasonic color science is dismal, I mean, gross greenish yuck, you won't want that. I mean, look how lovely that C300 frame is, that's pro gear and the rest are hobby toys, including the clunky BMCC. Consider avoiding 2000+ quid of disappointing "bargains" a 2000+ quid discount on something that's actually going to work. You're right... thanks for the advice! Whilst we're on the topic, are you able to advise me on a camera that would cut fairly nicely with the C100, for under £1000. It's for events videography... sorry to hijack! Quote Link to comment Share on other sites More sharing options...
jojo Posted May 23, 2013 Share Posted May 23, 2013 the biggest issue for me is the zig zag verticle line issue, which ive noticed on several videos already, if not most. Even if it is only an artifact due to vimeo compression, its worrying because thats were most content will end up. Is there a workaround for this? Quote Link to comment Share on other sites More sharing options...
jasonmillard81 Posted May 23, 2013 Share Posted May 23, 2013 I can respect the claims that a GH2/GH3 "can't touch raw" but why/how? In the two clips i've seen between the two it is not that noticeable to the average human eye. Unless you're pushing it to 400% (why I don't know) it doesn't look to be all that different. In other words, for the small gain you're getting over the gh3 is it worth the extra money on a body, storage (internal to the body), and extra hard drives you'll need, not to mention time to process, convert, grade, etc? I'm just curious as I know it is an individual basis. I'm curious if it is better to get a GH3 and add on a Metabones speed booster or get a 5D3 and use the h.264 footage the majority of the time and raw for more intense close-ups and landscapes. It's a trade off either way. Not sure how one would make a decision in either direction. Quote Link to comment Share on other sites More sharing options...
nahua Posted May 23, 2013 Share Posted May 23, 2013 the biggest issue for me is the zig zag verticle line issue, which ive noticed on several videos already, if not most. Even if it is only an artifact due to vimeo compression, its worrying because thats were most content will end up. Is there a workaround for this? The latest May 22 build doesn't have this anymore. Pattern noise is still there, though, if you lift the shadows up too much. Quote Link to comment Share on other sites More sharing options...
Administrators Andrew Reid Posted May 23, 2013 Author Administrators Share Posted May 23, 2013 I got to disagree here, sorry. Particularly on the comparison of the 1DC to the MKIII. We have both and use them extensively. The 5DMKIII's sensor was measured by DXOMark at 11.7 stops total. The 1DC gets around 12.5 in Canon Log at ISO400. It's not possible that the new 5DMKIII RAW has more dynamic range than the sensor is capable of seeing. I suggest you get a dynamic range chart before posting that sort of claim. :S DXOMark also measured the 1D C (1D X sensor) at 11.8 stops total so something is wrong with that Canon LOG 12.5 stop test. You said yourself you can't get more dynamic range than the sensor provides in the raw data! There's also a fundamental error of logic in the way you're comparing raw to MJPEG. You are suggesting that from the 11.7/11.8 stop raw 14bit sensor data, the 14bit raw video contains less dynamic range than compressed 8bit JPEG with LOG curve. Sorry but that makes absolutely no sense at all. By definition the raw data has the optimal dynamic range, and no matter what curve you apply to the JPEG, you lose quality when you compress it to 8bit. Quote Link to comment Share on other sites More sharing options...
Administrators Andrew Reid Posted May 23, 2013 Author Administrators Share Posted May 23, 2013 The latest May 22 build doesn't have this anymore. Pattern noise is still there, though, if you lift the shadows up too much. Was the fix for this made public? Just curious to know what caused it, whether it was in the sensor readout or in the debayering (Raw2DNG). Cheers Quote Link to comment Share on other sites More sharing options...
saulmarti Posted May 23, 2013 Share Posted May 23, 2013 Excellent article. Quote Link to comment Share on other sites More sharing options...
nahua Posted May 23, 2013 Share Posted May 23, 2013 Was the fix for this made public? Just curious to know what caused it, whether it was in the sensor readout or in the debayering (Raw2DNG). Cheers They enabled the bad pixel filtering, that processes and reduces any dead pixels on the sensor. So this helps the image and has removed the vertical lines from earlier builds. The Fixed Pattern Noise however is the sensor noise and that has yet to be worked on. Andrew Reid 1 Quote Link to comment Share on other sites More sharing options...
Max Posted May 23, 2013 Share Posted May 23, 2013 "The 1D C is of course top dog for resolution" Why? The 5d shoot still raw. Still has a resolution of a photo which is greater then compress video Quote Link to comment Share on other sites More sharing options...
Leica50mm Posted May 23, 2013 Share Posted May 23, 2013 I can respect the claims that a GH2/GH3 "can't touch raw" but why/how? In the two clips i've seen between the two it is not that noticeable to the average human eye. Unless you're pushing it to 400% (why I don't know) it doesn't look to be all that different. In other words, for the small gain you're getting over the gh3 is it worth the extra money on a body, storage (internal to the body), and extra hard drives you'll need, not to mention time to process, convert, grade, etc? I'm just curious as I know it is an individual basis. I'm curious if it is better to get a GH3 and add on a Metabones speed booster or get a 5D3 and use the h.264 footage the majority of the time and raw for more intense close-ups and landscapes. It's a trade off either way. Not sure how one would make a decision in either direction. I don't have a GH3, but the footage i have seen from it looks really nice. The color seems much better . Here's a video by Nick Driftwood : https://vimeo.com/54991683 Sorry. Quote Link to comment Share on other sites More sharing options...
peederj Posted May 23, 2013 Share Posted May 23, 2013 You're right... thanks for the advice! Whilst we're on the topic, are you able to advise me on a camera that would cut fairly nicely with the C100, for under £1000. It's for events videography... sorry to hijack! There is none. That camera is an anomaly...should be twice its current price, even though Internet denizens continue to insist it's somehow overpriced. The only thing cheap about it is the codec, which the $700 Ninja 2 solves. (OK and the EVF, which e.g. the Zacuto solves, but I can manage with the onboard one OK in a pinch.) Just stick with your 7D and use Cinestyle or another flat gamma of your choice and wait until you can afford a second C100. For events shooting even the most hostile critics agree the C100 is the best camera in the world. Quote Link to comment Share on other sites More sharing options...
peederj Posted May 23, 2013 Share Posted May 23, 2013 DXOMark also measured the 1D C (1D X sensor) at 11.8 stops total so something is wrong with that Canon LOG 12.5 stop test. You said yourself you can't get more dynamic range than the sensor provides in the raw data! There's also a fundamental error of logic in the way you're comparing raw to MJPEG. You are suggesting that from the 11.7/11.8 stop raw 14bit sensor data, the 14bit raw video contains less dynamic range than compressed 8bit JPEG with LOG curve. Sorry but that makes absolutely no sense at all. By definition the raw data has the optimal dynamic range, and no matter what curve you apply to the JPEG, you lose quality when you compress it to 8bit. The gamma can map any dynamic range to any bit space. You can have 20 stops mapped to 4 bits. However in that case when applying the S-curve to de-log it you will have awful banding, because there is reduced color fidelity with such a severe gamma. The 14 bits deep linear space the 5D3 is using gives it no more or less dynamic range than the sensor has and all of the color fidelity of the sensor. Using the log gamma does not change the dynamic range represented. 8 bits holds about 11 stops in a standard sRGB gamma, in a log gamma it can easily hold the full 12-13 stops of the sensor. However the color fidelity will be reduced a bit because information on fine gradations of level are being thrown out. So yes 14-bit linear RAW will be more accurate and less susceptible to banding than 8-bit Canon Log, but they are both likely to offer the full dynamic range of the sensor once mapped to a standard sRGB gamma for viewing. You will have slightly better color accuracy with the RAW. Nothing that Vimeo compression is likely to preserve for a natural grade, but something a severe stylized grade might reveal. Neat Video tends to smooth out the 8 bit banding in those cases very well btw. Quote Link to comment Share on other sites More sharing options...
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.