Popular Post Juan Melara Posted February 20, 2020 Popular Post Share Posted February 20, 2020 Hey guys, late last year I bought a P6K to work alongside the Alexa I normally shoot with. The goal was to have smaller, lighter camera that would live on a gimbal or could be used in situations where a full size Alexa wouldn't work. Knowing that both cameras would need to intercut seamlessly, I set about profiling the P6K using the same process I had previously used to profile the Alexa sensor and various negative film stocks. This process generates approximately 14,000 data points per camera. I then took the visual form of this data into Resolve and created a PowerGrade that accurately matches the P6K to the Alexa. After several requests, I've now packaged up the PowerGrade and added it to my store. You can check it out here: https://juanmelara.com.au/products/bmpcc-6k-to-alexa-powergrade-and-luts The PowerGrade At the heart of a PowerGrade is a custom 3x3 matrix that aligns the P6K's colours with the Alexa's. Because this is done through a matrix in linear space, the result is actually closer to a technical transform. The benefits this brings over a LUT is that it's non destructive, there's no danger of clipping or clamping data. And if you really wanted to, you could fully reverse the transform with zero loss in quality. This is because it doesn't break Resolve's 32bit float space like LUTs do. Probably one of the best features of a PowerGrade is that is fully editable and customisable. But since editing a 3x3 matrix isn't the easiest thing in the world, I've also included a Hue vs Hue/Hue vs Sat curve approximation of the matrix which is far easier to customise. And because it's a PowerGrade you can see exactly what is being done by each node, which means that I'm not able to hide snake oil like you can with a LUT. The LUTs Speaking of LUTs, knowing that a lot of people actually prefer to work with LUTs, I also created a LUT version. It's the exact same transform but in LUT form. One of the biggest limitations with most LUTs is that they clip any information that exceeds the top of the waveform, so basically any value above 1.0. This is a big issue with BMPCC footage as enabling Highlight Recovery or working with any ISO above 400 pushes valuable information above 1.0, where it is permanently clipped. With this BMPCC specific requirement in mind, I developed the LUTs to accept and work with values exceeding 1.0. So no matter if you enable Highlight Recovery or push the ISO to 6400, the LUT is able access and process all the information extending above 1.0. Anyway, check it out and let me know if you have any questions or comments. I'm keen to get feedback from any users to see what can be improved or expanded on. Comparisons Here are a few comparison images shot with the P6K mounted on the Alexa's top handle. The Alexa was using a Zeiss Milvus 35mm, the P6K was using a Zeiss Milvus 25mm, then cropped in post to match the FOV. The Alexa images are untouched, the P6K images feature minor exposure adjustments only. Check the store link for more images. Katrikura, majoraxis, Emanuel and 17 others 14 6 Quote Link to comment Share on other sites More sharing options...
anonim Posted February 20, 2020 Share Posted February 20, 2020 Compared pictures look to me identical, maybe a touch bit more of DR on P6K2Alexa part because of higher black point, judging from leaves in background and clothes. Although it is obvious one, I'd like to mention idea and give a vote for making similar PowerGrade transform also for P4K. Quote Link to comment Share on other sites More sharing options...
Attila Bakos Posted February 20, 2020 Share Posted February 20, 2020 Nice work, although 14000 data points sounds like overkill to verify a 3x3 matrix Quote Link to comment Share on other sites More sharing options...
Sage Posted February 20, 2020 Share Posted February 20, 2020 Nice work! heart0less and deezid 1 1 Quote Link to comment Share on other sites More sharing options...
kye Posted February 20, 2020 Share Posted February 20, 2020 Great stuff! Anything that looks that close from a side-by-side image comparison should be indistinguishable in any other situation.. Quote Link to comment Share on other sites More sharing options...
Juan Melara Posted February 21, 2020 Author Share Posted February 21, 2020 15 hours ago, Attila Bakos said: Nice work, although 14000 data points sounds like overkill to verify a 3x3 matrix Yeah you're correct, but I already had a library of those charts from various sensors and stocks, so it made sense to keep adding to it. Plus it's also actually quite helpful to have so many samples within the colour space, rather than just a few samples right at the edges. 15 hours ago, Sage said: Nice work! Thanks mate! Much appreciated. Another interesting thing I found in testing is how much dynamic range the P6K actually has. According to the charts published by BMD and ARRI, when both cameras are exposed at ISO800, the Alexa has 1 stop extra range in the highlights, but approximately the same amount of stops in the shadows. In testing I found this to be pretty much spot on. I found I could expose one of the charts exactly one stop higher on the Alexa before it touched the clip point. The interesting thing is how that actually translates in real world scenarios, and what that difference looks like. Here are some crops of a scene shot on both cameras. In LogC, so the differences are visible. I matched exposure on the day using false colour, they're as close as I could get them. Probably less than 1/20th of a stop of difference in exposure. The left image is the P6K. When comparing it to the Alexa image in the middle, thats exactly what 1 stop difference in the highlights looks like in a real world situation. It's interesting to see what happens to that 1 stop difference when highlight recovery is enabled – the P6K actually retains more highlight detail. Obviously you would never actually place important image information in that range (not that you would ever do that with the Alexa either). But if you need it create a smoother roll-off in anything neutral in colour like specular highlights, clouds etc, it is possible to outperform the Alexa in highlight dynamic range... sometimes. heart0less, majoraxis, cam1982 and 2 others 5 Quote Link to comment Share on other sites More sharing options...
thebrothersthre3 Posted February 21, 2020 Share Posted February 21, 2020 11 minutes ago, Juan Melara said: The left image is the P6K. When comparing it to the Alexa image in the middle, thats exactly what 1 stop difference in the highlights looks like in a real world situation. It's interesting to see what happens to that 1 stop difference when highlight recovery is enabled – the P6K actually retains more highlight detail. Obviously you would never actually place important image information in that range (not that you would ever do that with the Alexa either). But if you need it create a smoother roll-off in anything neutral in colour like specular highlights, clouds etc, it is possible to outperform the Alexa in highlight dynamic range... sometimes. That info was blown on the Alexa? Was the Alexa shooting RAW tho? Quote Link to comment Share on other sites More sharing options...
Juan Melara Posted February 21, 2020 Author Share Posted February 21, 2020 5 minutes ago, thebrothersthre3 said: That info was blown on the Alexa? Was the Alexa shooting RAW tho? Yeah that info was gone on the Alexa. The Alexa was shooting Prores 444, but there is no difference in dynamic range between Prores 444 and ARRIRAW. And there is no highlight recovery on ARRIRAW either, so there would be no chance of bringing it back. Quote Link to comment Share on other sites More sharing options...
thebrothersthre3 Posted February 21, 2020 Share Posted February 21, 2020 3 minutes ago, Juan Melara said: Yeah that info was gone on the Alexa. The Alexa was shooting Prores 444, but there is no difference in dynamic range between Prores 444 and ARRIRAW. And there is no highlight recovery on ARRIRAW either, so there would be no chance of bringing it back. That is crazy honestly. According to cinema 5d the Dynamic range on the Pocket 6k is 11.5 stops. With 1 stop in highlight recovery that is 12.5 stops. So the other 1.5 is in the shadows on the Alexa? Or does the Pocket 6k actually have like 12.5 stops of dynamic range even before highlight recovery? Quote Link to comment Share on other sites More sharing options...
Juan Melara Posted February 21, 2020 Author Share Posted February 21, 2020 22 minutes ago, thebrothersthre3 said: That is crazy honestly. According to cinema 5d the Dynamic range on the Pocket 6k is 11.5 stops. With 1 stop in highlight recovery that is 12.5 stops. So the other 1.5 is in the shadows on the Alexa? Or does the Pocket 6k actually have like 12.5 stops of dynamic range even before highlight recovery? The Cinema 5D guys measured at ISO400. Rating and exposing at ISO800 shifts the mid point, allocating one stop extra to the highlights. It also lifts the shadows, revealing one more stop down the bottom. That last stop will be slightly noisy. And thats what can make measuring DR tricky, as each tester will have a different tolerance to what is an acceptable level noise and whether the bottom stops actually count towards usable dynamic range. And if you're Red, just count everything and say it's 17+ stops. Here is another image to show you the bottom stops of both cameras The previous example didn't have deep enough shadows to crop into. The exposure is matched on the day with false colour, this time I reckon they're identical. No noise reduction. Shot BRAW 5:1 / Prores 444 on the Alexa, which would be compressing out a bit of noise. Remember that the P6K has actually been cropped in to match the FOV of the Alexa, as I shot a 25mm on P6K and a 35mm on the Alexa. The crop is 1.32, so I'm actually only using a 4.6K portion of sensor. It would be less noise if I used the full sensor and scaled down. majoraxis and heart0less 2 Quote Link to comment Share on other sites More sharing options...
Juan Melara Posted March 2, 2020 Author Share Posted March 2, 2020 After many requests, P4K2Alexa PowerGrade + LUTs coming in the next day or two. Dawn to dusk, one PowerGrade, no adjustments to the transform. Katrikura, kaylee, cam1982 and 5 others 5 3 Quote Link to comment Share on other sites More sharing options...
MoroZ Posted March 2, 2020 Share Posted March 2, 2020 I purchased the p6k version for use with p4k. Is a discount possible? techie 1 Quote Link to comment Share on other sites More sharing options...
Geoff CB Posted March 2, 2020 Share Posted March 2, 2020 You get damn close here, great job. The highlights in the Alexa just cannot be topped. Quote Link to comment Share on other sites More sharing options...
heart0less Posted March 2, 2020 Share Posted March 2, 2020 Cool! Is there any chance for a X-T3 / X-T4 version, as well? Quote Link to comment Share on other sites More sharing options...
Simon Young Posted March 2, 2020 Share Posted March 2, 2020 @Juan Melara I know you probably get this all the time but you’re a genius and a prophet. heart0less and zerocool22 2 Quote Link to comment Share on other sites More sharing options...
Juan Melara Posted March 3, 2020 Author Share Posted March 3, 2020 P4K2Alexa is out. Accurately match BMPCC 4K footage to the ARRI Alexa via a clean, non destructive 3x3 matrix. With none of the downsides of using a LUT. Fully editable, no clipping, no clamping, no snake oil. https://juanmelara.com.au/products/bmpcc-4k-to-alexa-powergrade-and-luts heart0less, techie, BopBill and 1 other 2 2 Quote Link to comment Share on other sites More sharing options...
Juan Melara Posted March 3, 2020 Author Share Posted March 3, 2020 23 hours ago, MoroZ said: I purchased the p6k version for use with p4k. Is a discount possible? I was going to send it to you for free, but then you wouldn't appear on the customer list, so you wouldn't get any free updates. Flick me an email and I'll get you a full refund or figure something out. You definitely want this version as the sensors are quite different. 23 hours ago, heart0less said: Cool! Is there any chance for a X-T3 / X-T4 version, as well? I could make one. Just need to see what the demand is like and then would need to source the cameras. 23 hours ago, Geoff CB said: You get damn close here, great job. The highlights in the Alexa just cannot be topped. Yeah there's a good stop and a half extra on the Alexa. Highlight recovery brings that down to maybe a stop difference. I'm glad you didn't mention the Alexa's "highlight roll-off." The Alexa doesn't have a highlight roll-off! It's completely linear up there. heart0less and techie 1 1 Quote Link to comment Share on other sites More sharing options...
heart0less Posted March 3, 2020 Share Posted March 3, 2020 50 minutes ago, Juan Melara said: I could make one. Just need to see what the demand is like and then would need to source the cameras. That's a great news. I think the original X-T3 is highly underrated camera, but seeing how X-T4 is welcomed, it may no longer be the case in the upcoming months. Luckily, both of them, apart from sharing the same sensor, seem to have the same color science. They don't offer RAW, unfortunately, but 10-bit @400Mbps is still plenty of data for color manipulation. persol and techie 2 Quote Link to comment Share on other sites More sharing options...
MoroZ Posted March 3, 2020 Share Posted March 3, 2020 I am using P6KPowerGrade at bmpcc4k. is there a big difference if I use P4KPowergrade, or is the difference between them minimal? And one more question: how is it going with cutting off the red channel at P4KPowergrade? Quote Link to comment Share on other sites More sharing options...
Juan Melara Posted March 3, 2020 Author Share Posted March 3, 2020 5 minutes ago, MoroZ said: I am using P6KPowerGrade at bmpcc4k. is there a big difference if I use P4KPowergrade, or is the difference between them minimal? And one more question: how is it going with cutting off the red channel at P4KPowergrade? The sensors are different enough that I wouldn't even think they are part of the same family. I actually wasn't expecting that. It required a bit of work to even out the non linear quirks. As long as you process the footage at 0 tint, P4K2Alexa will sort out the red channel, well all channels actually. 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.