-
Posts
1,839 -
Joined
-
Last visited
Content Type
Profiles
Forums
Articles
Everything posted by jcs
-
The 840fps footage looks lower noise than the 960fps on the FS700, though seems to be some kind of fixed pattern artifacts from binning/scaling? 4K/4000fps options are for a future product? 4.7 micron vs A7S 8.5 micron pixels with similar low light performance? A nice start and very low price for what it can do.
-
Gale Tattersall was the DP- he was also the DP for House M.D., and used 5D's on the show: http://www.eoshd.com/2011/02/eoshd-interview-with-house-director-of-photography-gale-tattersall/ . It started a little weak but got better as it went on. As a new camera test/demo it's fine; perhaps had a low budget and had to be rushed for NAB. The C300.2 will get used for lots of Netflix level work, and the production, acting, etc., was better than a lot of content on Netflix. Reverie looked amazing and had cool music, but no dialog: a music video (much easier than a narrative). Hard to find a 1080p version of it: https://vimeo.com/7151244 , http://www.learn.usa.canon.com/galleries/galleries/sample_videos/reverie.shtml .
-
By shooting color charts with your camera and a Canon you can create a transfer function in the form of a 3D LUT. This could be possible with something like SpeedGrade (and/or Resolve if it has a frame/color match feature). This is tricky as 3D LUTs are sensitive to exposure, color temperature, etc. The best way to get Canon (or better) skin tones is to shoot with ARRI (). I shoot stills or raw video with the 5D3 under the same conditions as the A7S to then match the A7S to 5D3 footage in post (using white balance, curves, ProcAmp, etc. in Premiere). The next version of Premiere has a color style match feature from photos- that might also help (shoot Canon/5D3 stills to provide the color match template).
-
For white/blonde hair a blue screen can be easier to key. To speed up ProRes 444 with Photon, try setting the Options/ProRes QScale to 6 (to start) and compare time/quality/size while adjusting QScale up/down.
-
I've been experimenting with creating 422 10-bit H.264 (ALL-I and IPB) using Photon as well as editing the files in Premiere. While the files work for the most part, the performance wasn't as high as I would expect. 444 10-bit H.264 didn't work at all. Thus, even though Canon's XF-AVC is part of the H.264 standard, NLEs may need to update their H.264 decoders to handle more parts of the spec.
-
Sensor readout is faster: resolution is irrelevant - RS is reduced per Canon.
-
Thanks :). Yea a little more red & blue (magenta) would help. Interestingly, while the face has a bit too much green, other areas have too little green (pinkish/magenta). If short on time (though with Premiere Pro masking, it's pretty fast, especially if the camera/actor aren't moving much), it's best to correct for the face (and let other skintones go more magenta (this case)). The 5D3 seems to do a good job keeping skin out of the less-pleasing green/pink zones. Figured it was part of Canon's color science.
-
What camera was used to shoot this? Watching on an iPhone 5S (Retina- generally excellent color), the face skintones look too green. Don't tell me he's an alien and the color is accurate. Everyone knows the green aliens top out around 4'.
-
It's unlikely to have IQ matching or exceeding Alexa or F65- if it gets close, it will be a pretty good deal. High quality accurate face tracking autofocus with EF lenses and assisted manual focus with all lenses is the killer feature exclusive to this camera. If it works well, it's a very important feature not available with any other camera in this class.
-
Another H.264 variant for recording internally: http://www.canon.com/news/2015/apr08e.html. Basically Canon's version of Sony's XAVC (both of which are already defined in the H.264 spec.). Wouldn't be surprised if they licensed the code from the same software developer. 422 10-bit Intra (ALL-I) is very similar to ProRes, likely more efficient. They also support long GOP (interframe/IPB vs. intra) for 2K and below, up to 12-bits, which is very efficient. 10-bit H.264 ALL-I & IPB recorded internally is excellent.
-
Canon has variable analog gain on the sensor. It appears the other chips have fixed analog gain which is optimized for the best noise performance.
-
SleepyWill- what issues are you having with the Pre2? PC or Mac? IIRC, the issues were on the PC (ASIO driver); it's been a while since using it with a computer (vs. studio camera preamp. It can also work well in the field with a USB battery). For PC/Mac recording the Scarlet 2i2 is a very good deal: http://www.amazon.com/gp/aw/d/B005OZE9SA/ for $150. It used to have an occasional driver issue after coming out of sleep, haven't seen it in a while after MS updates (I didn't change the drivers). I had the RME Fireface 800. Incredibly stable drivers- had zero issues. TotalMix etc. is way too complicated- always had to spend time figuring out routing if I hadn't used it to record in a while. I had purchased the Scarlett 2i2 for a friend to use for VO. I tested it first and found it sounded better than the FF800! For desktop work 2 channels is plenty, so I sold the FF800 and replaced it with a Scarlett 2i2. Sound Devices preamps and limiters are stellar- if you want the best sound and to also be able to handle surprises in peak volume with no worries, SD gear is worth it. That said, for VO/desktop recording, Focusrite Scarlett series is good enough and a great value. To get 'better' sound than either of these preamps, you've got to spend a lot more, and by better it's not so much less noise but different character. Spend time on Gearslutz for tons of info. Grace makes excellent clean mic preamps. Here's a poll: https://www.gearslutz.com/board/high-end/673657-whats-your-favorite-clean-preamp.html With audio there's tons of esoteric gear that each owner swears by. Most of the gear colors sound. I used to think- record pristine and tweak/color in post with filters. Now I appreciate colored sound for efficiency- similar to getting the look you want in camera. The main place to color the sound is with the mic. Ribbon mics for example sound incredibly smooth with a bass emphasis (need a clean 60dB capable mic preamp (Scarlet and Pre2 work)).
-
Thanks mercer. Elena (main actress: http://www.imdb.com/name/nm3025841/ ) is in Furious 7, just released.
-
Sophisticated focus assist technology- fully automatic with tracking as well as hybrid manual modes. Current autofocus and manual focus assist tech is very primitive compared to what's possible now: manufacturers have been holding back. Dual pixel sensors and/or IR, ultrasonic, laser, etc. for 3D depth-finding to not only provide better focus but to also show the focus plane in real-time in the display (depth-map rendered etc.).
-
The GH4 can do a bit better with DR by using iDynamic to bring up the shadows, then underexposing to protect the highlights: http://***URL removed***/previews/panasonic-dmc-gh4-sony-alpha-7s/14
-
mercer- yes, thanks, Delta was our first narrative short. We had no complaints for audio/sound (other elements- yes; a learning experience for us! ). Glad you got the humor shots! (I shot everything- no stock footage). The short was supposed to be campy and fun; special effects a homage to old-school video games / Star Wars, etc. We'll improve the story (and everything else) on the next one The simplest solution would be a shotgun mounted on camera with a high-quality isolated mount (and far enough away from the lens to mask stabilizer noise, etc., when present), with a decent preamp (either in the mic system (Rode etc.)), or as another piece of hardware (JuicedLink sounds cleanest for the money, SD's MixPre-D sounds much better and has superior limiters: worth over double the cost IMO. The hacked iRig Pre is perhaps the best low cost, ultra small solution: http://www.dslrfilmnoob.com/2012/11/25/irig-pre-hack-cheap-xlr-phantom-power-preamp-dslr/ ). You could also rig a boom to a backpack, etc., to get the mic closer to the subject (really need to get mic pointed down towards the ground to take advantage of noise rejection). Regarding dual sound, IMO it's only worth it for larger productions, with a dedicated sound guy, and only when using Sound Devices or similar quality gear (I have a Zoom H4n and Tascam DR100mkII: not high enough quality vs. internal DSLR with a preamp to warrant extra effort of separate sound. DR680 and newer are good and SD 702 and up are preferred. SD gear has amazing preamps, not only clean, but a very full, natural 'Hollywood' sound. The SD limiters are also very, very good: the extra cost for SD gear is worth it: http://www.bhphotovideo.com/c/product/429566-REG/Sound_Devices_702_702_High_Resolution_2_Channel.html . How many 5.0 reviews do you see on ANY piece of gear? Pretty amazing). Steve M.- going wired like that can certainly sound better than wireless (until getting to the Zaxcom or Lectrosonics digital wireless level), though the Sennheiser (and similar priced gear) is more than good enough for indie work (even material planned for streaming paid delivery). Another solution that can be 'free' is to use old cell phones as lav recorders (even your current cellphone). Rode makes a decent lav for the iPhone & Android: http://www.amazon.com/Rode-smartLav-Lavalier-Microphone-Smartphones/dp/B00EO4A7L0 . It is technically possible to have an app be remotely controlled to start recording, meaning a bunch of iPhones/Androids could join an adhoc wifi network and be triggered to record remotely, then send a compressed AAC copy over wifi back to the controller, where mixed audio could be monitored live (if this already exists- very cool)). In post the locally recorded uncompressed WAV files can be used for editing (along with timecode and/or sidecar metadata to make syncing easier). Not monitoring lav recordings live is indeed very risky- many times the mic/cable rub and must be readjusted due to talent movement (cable loop taping and careful placement help, but there are still issues that come up during recording).
-
Tempted? Should you get a $6499 Canon C300 or wait until after NAB?
jcs replied to Andrew Reid's topic in Cameras
In Hollywood, looks do matter. For everything. Fortunately Luc Besson & his DP gave the F65 a chance and Lucy looks fantastic (as does Oblivion). Here's the quote from Luc's DP, Thierry Arbogast, AFC: Full article- excellent read! http://community.sony.com/sony/attachments/sony/large-sensor-camera-F65/523/2/Lucy_Special_Report.pdf They also used Alexas (steadicam) and Reds (multicamera car cam)... as B-cameras! Even the 5D was used for a few shots. -
The trick to getting professional audio (low noise, excellent quality) is to use a preamp into the DSLR. In the field, the Sennheiser G3 wireless lav system works great, especially for run & gun and guerrilla shooting when you don't (or can't )have a boom operator. I use two G3 systems for two channels and a simple Y-adapter into the Sony A7S and it works great. In the studio / green screen I use a Sound Devices USB Pre2 (also a stand alone preamp, same hardware topology as the 744T) and an Audio Technica BP4029 mid-side stereo shotgun. For ADR in post I use a Shure SM7 mic into a Focusrite 2i2 (amazing quality preamps for the price. Not as good as Sound Devices, but it's a more stable solution as a computer input device (SD made some odd choices for the USBPre2 hardware (as in it never turns off, even when the computer sleeps + some driver issues)). Example audio: http://brightland.com/w/delta/ : jump to 6:30 to hear VO (SM7 + Scarlet) and BP4029 + Sound Devices (into a GH4) for prelude to fight scene. Earlier scenes used the G3's into the A7S. All audio sweetening done from within Premiere Pro (used to use Pro Tools, including their crazy expensive hardware) and small amounts of Audition (waveform editing). For shorts/indies/corporate/for-fun, this level of hardware is plenty good!
-
Tempted? Should you get a $6499 Canon C300 or wait until after NAB?
jcs replied to Andrew Reid's topic in Cameras
The C300 was (is) popular because it provided fantastic quality with little effort. That's Policar's point regarding working pros choosing it so often, and why it was one of the top rented cameras. By using 5D3 RAW as a reference, I was able to figure out how to make the A7S look good (biggest challenge was skin tones). That took time & effort- I and most people would prefer to spend time shooting and editing (creating) vs. figuring out something that should work 'out of the box', which is what Canon does very well. Sony stepped up their game with the FS7, but it sounds like the firmware is still a mess. Canon releases solid cameras from day one. Regarding color, I agree the Fuji X-T1 is on par and even better than Canon. Makes sense, since Fuji is a color company (lots of film experience). However, when helping a non-techie friend buy a new stills camera, as soon as I saw the menu system of the X-T1, I realized it was an instant no-go for her. She liked the smaller size and image quality, but the camera was way too complicated. The 5D3 is far simpler and easier to use (which is what she chose- the latest pricing made it an easy decision). In terms of absolute best color & resolution with little or no artifacts, the F65 is on the top of my list. Reportedly the reason it's not more popular is the camera itself is ugly. I suspect the file sizes and workflow might also be challenging/expensive. So ARRI still has the best balance, though it sounds like Canon is stepping up their game for NAB to compete with ARRI. Guessing the Varicam 35 isn't used more due to price and, if the F65 'ugly not used' report is true, could be Panasonic doesn't have the prestige to get people to switch from ARRI. So it would appear that Panasonic would have to drop the price (at least for a while) to get converts and buzz. Color & resolution are good enough now. DR (including highlight behavior) and noise (low light) still have a ways to go, as does more affordable HFR. However, what really needs work is autofocus and more importantly, intelligent focus assist. That is, manual focus, but with 3D depth technology (IR, ultrasonic, laser, photogrammetry, 'focus/depth pixels' etc.). Where the camera can accurately display a 3D slice of the focus plane rendered into a depth-mapped 3D scene (the viewfinder could even be stereo 3D as an option, but this would work without stereo3D). Embedded computers & GPUs are easily fast enough to do object tracking, which can further assist with focus, where the operator can move the focus track point(s) as well as focus depth plane in real-time while recording. Accurate dynamic (non-rehearsed) manual focus, even for experienced full-time focus pullers for 4K and beyond is very challenging. With budgets constantly shrinking, providing a reliable means to perform accurate focus by the camera operator is long overdue. -
Magic Lantern run Linux 3.19 on Canon DSLRs - download source code
jcs replied to Andrew Reid's topic in Cameras
This A completely custom GUI to run the camera- with complete control over everything: not running any of Canon's firmware would be possible. Instead of hacking on top of existing firmware/OS, now it's possible to completely take over: potentially higher performance, new functions/features. Loading codecs- not likely however if there's any way to access the custom hardware to perform debayering at a higher quality, then sending to the custom hardware for H.264 encoding at a higher bitrate, it could be possible to have RAW-like resolution with H.264. If possible to access the debayer hardware as well as individual elements of H.264 hardware, such as the DCT, it could be possible to create something compatible with ProRes (not likely the ARM processor can do everything by itself). -
While there's no difference in the brightest and darkest pixels in 0-255 vs. 16-235, there are only 220 code values for 16-235 vs. 256 code values for 0-255. 36 code values aren't a huge difference, but could affect banding in some cases (less contrast range).
-
Photon doesn't do anything special between ProRes 1 and 2: default FFMPEG behavior. Here are some Photon Options / Custom Options for FFMPEG to play with: -vf colormatrix=bt709:bt601 -vf colormatrix=bt601:bt709 -color_primaries bt709 -color_primaries bt470bg (I didn't see any change with -color_primaries for the cases I tested) -vf scale=in_range=full:in_color_matrix=bt709:out_range=full:out_color_matrix=bt709 -vf scale=in_color_matrix=bt709:out_color_matrix=bt709 bt709 can be swapped with bt601 for above args. I would expect the -vf scale color in_/out_ matrix above to work the same as the -vf colormatrix compact form further above. -vf scale=in_range=pc:out_range=pc pc arg above can be: pc (0-255, full range. 'full' is also valid) or tv (16-235) If anyone finds/figures out custom FFMPEG args to fix any issues with ProRes, please let me know.
-
They appear to be using FFMPEG elements for file reading (LGPL libraries vs. FFMPEG itself) and appear to be using AVFoundation (etc.) which is native OSX if using Apple's ProRes for final output. At which point AVFoundation supports H.265 decoding (does it already?), FFMPEG (and libraries) wouldn't be needed at all to transcode H.265 into ProRes. AVFoundation is super easy to code for (especially compared to DirectShow on Windows). If AVFoundation already supports H.265, a custom tool for H.265 to ProRes would be fairly quick & easy to create (a few hours of coding- GUI, etc. would take more time). MainConcept also has decent tools/libraries (possibly much faster than the free/GPL/LGPL stuff), though the licensing cost would up product cost.
-
Thanks guys. Geoff- glad to see the colors/contrast are working well for you. As noted in the Photon manual, FFMPEG introduces a color shift for 4K material with ProRes. Not a major deal if all files are transcoded the same way. However, I prefer to have color/contrast preserved and hopefully FFMPEG devs can fix the color shift issues with >1080p material in the future. In Photon under Options/Custom Options, you can try custom FFMPEG commands which adjust matrices (bt601/709) and input/output range (video/full range (16-235, 0-255, in and out)). I'll post examples shortly. I also suggest trying out high bitrate H.264. In Premiere Pro and FCPX, H.264 can edit fast enough on modern computers and GPUs, including 4K. 10-bit 422 H.264 requires building FFMPEG (link to instructions in Photon manual). Due to licensing issues, I can't host FFMPEG builds. 10-bit 422 H.264 (including ALL-I versions, which Photon supports) should be competitive to ProRes with smaller files (though will not be as fast for editing; however can be more than fast enough). 10-bit 422 H.264 is basically XAVC, which Sony provides in their pro cameras. When performance testing, try both ProRes Modes 1 & 2. Mode 1 is faster for smaller jobs-at-a-time, and Mode 2 is faster for many jobs-at-a-time. Mode 2 can be much faster when adjusting ProRes QScale (not available in Mode 1). Higher numbers are faster, resulting in smaller files with lower quality. Perhaps do just one reference file with QScale 1 (max quality possible), then try higher QScales and A/B test in your NLE at e.g. 400% to check for macroblock and other artifacts, then adjust QScale to get a balance of transcode time, quality, and file size. The current version of Photon limits the number of jobs based on CPU cores available. I can unlock this, however another issue is since I don't directly control FFMPEG (runs as a separate process), I can't strongly control memory usage. Photon monitors memory and CPU load, and adjusts total jobs in real-time. Under certain conditions, FFMPEG can (somewhat unpredictably) allocate a lot of memory, and can cause the computer to run out of physical memory, forcing the computer to page to disk, which is extremely slow (select File/Abort Current Jobs if this happens: I could do this automatically in a future build (when physical memory hits zero)). I have some algorithm ideas for Photon to work around this condition without aborting, but isn't something I can implement quickly. For now, set Max Jobs to prevent this issue if it ever happens (I prefer to have software be smart and 'do everything automatically' when possible). Photon's MB/s indicator is MB/s written to disk for all jobs. Sometimes the processing rate is disk-limited vs. CPU-limited. Fastest processing occurs with SSDs (especially on the output side).This is different than FFMPEG's reporting of bitrate (which is the video file's current bitrate). I will add FFMPEG's bitrate output in the next build of Photon: this can be helpful in tuning output file size. Thanks for the feedback so far and look forward to more benchmark numbers.