Sony XAVC-S in the HDR-CX405

[VidThreeNorth]

Where I Was:

I have had cameras and camcorders before, but around 2014-15 I started outfitting myself for 3D. So I needed a pair of matched camcorders. I chose a pair of Canon R40's. At the time I bought them, the R40's had a firmware upgrade bringing a new 1080p 30 fps file format. Without the upgrade the camcorder would record either 1080p 60 fps or 1080i 60 fps (which captures at 30 frames per second but saves the image it interleave pairs of 60 FIELDs per second. A "field" is essentially 1/2 of a frame, by taking alternating lines, so to get the full "1080" line image requires collecting the 2 FIELDs and recombining them into a single FRAME.

At that time, some people were having difficulty working with this "interlaced" data format. Today it is not really that much of a problem. So in that regard, it is difficult to say whether there was any significant change in the camera at all. I recall another firmware upgrade around that time that had to do with WiFi, so perhaps that was a "real" upgrade.

At about that time, Sony's competing camcorder was the CX240, but shortly after that, Sony replaced the CX240 with the CX405. This was much bigger upgrade, but since that upgrade, going forward, Sony has not made a further upgrade in that price bracket since.

There were two big changes in upgrading from the CX240 to the CX405. First, the new CX405 gained optical image stabilization (sensor shifting) where the CX240 had only digital stabilization. On top of that, the CX405 gained a new recording mode: XAVC-S 1080p 30 fps @ 50 mbps.

Until then, Canon's R40 and its predecessor had better image quality than the Sony -- for example, at 1080p 60 Frames per second the Sony stored around 24 mbps where the Canon stored around 34 fps. This superiority was at every video recording format, and if you made comparison videos the Canon generally looked better. The only exception was when light dropped down low and the Sony "back-illuminated" sensor would be better, until lighting was so bad that even the Sony would need supplemental lighting.

But with XAVC-S, Sony claimed superior image quality, which should surpass Canon. The problem with this claim is that Sony defined XAVC-S as a set of limits. And there was no guarantee that to top end of the potential specs were implemented in any particular camera. Moreover, in some cases, particularly for the CX405, Sony has not stated what the limits of its version of XAVC-S actually are.

For example, XAVC-S "can" be 10-bit colour. But it can also be 8-bit colour which is no better than standard "common" AVC video. Similarly, it can have YUV 422 chroma sub-sampling. But it might only have "common" YUV 420 chroma sub-sampling. This non-manditory nature of the spec means in effect that Sony could simply change the file package from its AVC type output and then call it "XAVC-S" with no improvement at all. Making some test files, I did confirm that the output is actually around 50 mbps, so at least it probably had more detail. Personally, I was hoping for 10-bit colour and YUV 422. The 10-bit colour probably won't help me much, but the YUV 422 chroma sub-sampling is a major improvement for me.

However, when I recorded test clips I was left with doubts. When played back using VLC, its data reports an ordinary AVC1 compressed stream with a YUV 420 decode format. Windows 8.1 does confirm a video bitrate near 50 kbps. The audio is disappointing PCM. But emphatically, so far I have no confirmation of either 10-bit colour or YUV 422 chroma sub-sampling.

VLC:
Stream 0
Codec: H264 - MPEG-4 AVC(part 10)(avc1)
Resolution 1920x1090
Frame rate: 29.970029
Decode format Planar 4:2:0 YUV [This correct?]
Stream 1
Type Audio
Codec PCM S16 BE (twos)
Channels Stereo
Sample rate 48000 Hz
Bits per sample 16

Win 8.1 Properties
Created Aug 29, 2018 16:53:43
Size 731,100,731 bytes
Length 01:53
Frame width 1920
Frame height 1080
Data rate 49,528 kbps
Total bitrate 51,059 kbps
Audio
Bid rate 1530 kbps
Channels 2 (stereo)
Audio sample rate 48 khz


Unfortunately, I do not have a properly set up pair of recordings of the same subject matter. I can only upload a pair of "nice" captures that confirm that both camcorders can do a "good job". Hopefully, I will eventually do a proper test recording of the same subject matter by both camcorders.

If anyone has seen a reliable report of what parameters are used for the CX405's version of XAVC-S, I would appreciate such information.

"01-0005-19h26m22s863-C1.jpg"
- a captured from from a 1080p Canon R70 video clip recorded using "standard" H.264 compression at 24 mbps. The PNG version was 2,459,668 bytes.

"01-C0002-14h33m07s657-C2.jpg"
- a captured frame from a 1080p Sony CX405 video clip recorded using its version of "XAVC-S" at 50 mbps. The PNG version was 3,383,721 bytes. NOTE: When I compressed the Sony frame to a C1 level JPEG the output file was too big to upload "here". Instead, I have uploaded a C2 level compressed version. So if you look at the files, keep in mind that the original captured frame was even slightly better than this. The JPEG "C1" compressed file was actually 2,116,429 bytes.

 

[VidThreeNorth]

HDR-CX405 Lens and Still Pictures:

I was not happy with the frame capture that I posted above. It looks softer than I expected. I was wondering if the lens was "borderline", which would not be a surprise considering the high zoom ratio and maximum F-stop. I decided to see what I could tell by taking still pictures with the camcorder. This was a new experience for me. I cannot recall ever taking a still picture with any of my camcorders. Why would I want to? I have always had good still cameras. And yes, the still pictures taken with the CX405 are not as good as I can get from my still cameras, but some of them turned out quite good. Aside from the lack of a "raw" file, the main issue I have in this case is the 9 M-Pixel resolution, which comes from the specialized (16:9) "designed - for - FullHD" sensor.

I have a few pictures I want to upload, but unfortunately, a new problem has come up: My mouse is failing. This has been a problem for a couple of weeks now, and it means that I cannot use my editing programs to select a crop area.

This brings up a deficiency that I have wanted to mention for years. All these graphics programs have overly specialized interfaces. There is no good reason not to allow a text numeric crop specification. Even when my mouse is working perfectly there are times when I know exactly what crop I want numerically and I am wasting time trying to mark out that crop with a mouse. But in this case where I cannot get my work done at all because of the hardware failure the program's design failure -- and that is what it is, means I have to wait for a replacement mouse to arrive. Specifically I am trying to use Corel PaintShop Pro 2018, but I cannot think of any other program that is better in this regard. Technically, the program design failure is called "lack of robustness".

Anyway, today I am uploading a single sample file that happens to be small enough out of the camera. This is mainly due to a generally "simple" subject matter. I will probably upload versions of the other files later. But overall, my first tests indicate an adequate lens.

I do not know the flower but the blooms seem to be about 6" in diameter. In this image I manually decreased the exposure to preserve the colours and details of the petals. The EXIF data does not indicate the manual exposure change.
[2018-09-04 I had a chance to measure the blooms with a tape measure and the larger ones are about 8" wide.]

"DSC00025.JPG" -- Pink flower 6" bloom
Partial EXIF:
2018-09-01 13:00
HDER-CX405 v1.00
Dimensions 4032 x 2272
Bit depth 24
Resolution unit 2
Color representation sRGB
Compression bits/pixel 1
F-stop f/3.7
Exposure time 1/500 sec.
ISO speed Nil [an odd discrepancy]
Exposure bias 0 step [manual]
Focal length 3 mm
Max aperture 2.2734375
Metering mode Pattern
Digital zoom 1

NOTE: the pictures from this set were all taken around 1 - 2 meters from their subject matter. I did not plan on making this a "close focus" test, it just worked out that way.

[2018-09-02 18:33]

"DSC00037.JPG" - [Not Uploaded]
- 2 monarch butterflies
- 1 narrow view and 1 at 3/4 view

This picture gives me confidence in the Sony - Zeiss lens. The focal length is almost the same as the previous picture but I am closer, and the exposure was right. I am not uploading the original file because a better starting point is seeing what the lens can do in a 1920 x 1080 image. This is not what you will get in a frame grab because the H.264 compression will remove some of the quality, but at least this confirms that the lens is not the limit for this camera -- at least not in this situation, and I think likely for most of the mid-range and long focal lengths.

I started by cropping the image down to what would be a "UHD" frame -- if the camcorder had the supporting electronics. I then resized from the 3840 wide UHD size down to 1920 x 1080 size. That image is:

"DSC00037b--rsz1920-C1.jpg"

The only negative that is evident is that I believe noise removal is more than I generally use, but it does not seem excessive. Sharpening is about right.

The second version is a "full size" detail crop.

"DSC00037c--Crop02-1640-C1.jpg"

This shows the theoretical performance of the lens if it were used for UHD (~4K) video. It is not really worth concerning oneself with this possibility since it appears that this sensor's noise would be a limit of such an upgrade, as well as the processing capability of the chipset.

Since the zoom range of the lens is 1.9 mm - 57.0 mm (35mm camera equivalent view 26.8 mm - 804.0 mm). At ~25.7 mm, this picture is near the end of the telephoto range. From here, I just need a wide-angle picture to have a good idea of its capabilities.


Partial EXIF:
Corel PaintShop Pro 2018
Software HDR-CX405 v 1.00
Date and time: Sept 1, 2018 13:15:19
Pixel height 2272
Pixel width 4032
Component configuration YCbCr
Color space sRGB
Exposure mode Auto exposure
Exposure bias 0.00 ev
Exposure time 1/180 sec.
F number f/3.4
Max aperture f/3.4
Focal length 25.7 mm
Metering mode Pattern
Digital zoom ration 1.0:1
NOTE: No ISO information

Per Win 8.1:
Size 2,545,579 bytes
Bit depth 24
Resolution unit 2
Colr representation sRGB
Compressed bits/pixel 1

[2018-09-04]

Wide Angle:

The "pink flower" was taken with a sufficiently wide focal length but the underexposure is not preferable for this analysis, so I am looking at:

"DSC00032.JPG"
[Not uploaded]
- black eye susans which used a focal length of 3.1, which is just a bit longer

Win 8.1:
Sept 1, 2018, 13:04:57
3,247,516 bytes

Partial EXIF:
Software: HDR-CX-405 v. 1.00
Pixel height 2272
Pixel width 4032
Component configuration YCbCr
Compression bits per 2.000000
Color space sRGB
Exposure Auto exposure
Exposure bias 0.00 ev
Exposure time 1/500 sec.
F number f/3.7
Max aperture f/2.4
Focal length 3.1
Metering mode Pattern
Digital zoom ratio 1.00:1


"DSC00032b-rsz1920-C2.jpg"
The camera file has been cropped to "UHD" equivalent frame, then resized down to "Full HD" image size. No alterations except for the crop and the resizing.

"DSC00032c-Crop02-1640-C1.jpg"
This is a detail crop from original camera file. No alterations except for the crop.

When resized to 1920 (Full HD) the resulting C1 compressed JPEG was about 2.237 MB, which is too large to upload, so I re-compressed it at C2. So again, keep in mind that the picture is was just a little bit better than the "1920" that I have uploaded. When I look at the detail crop and the resized "1920" file, again, the lens does not seem to be a problem.

But looking at the residual image noise implies to me that the real problem is probably the sensor and the processing. I would be more confident in what I am saying if I had a "raw" file to look at, but from looking at these stills, it seems to me that the sensor is producing a fairly noisy image, and that when the processing is reducing the noise, it is probably taking some of the detail with it. As I noted before, the pictures are not being "over sharpened", which is good, but it is leaving the result looking soft.


Temporary Conclusions:

What I see so far makes me think that the CX-405 will give a better "Full HD" video image than the CX-240 which it replaced (and in my case, it will eventually replace my current CX-240). However, I was hoping that it would "almost completely replace" my Canon R70 as well for 2D "Full HD" video work. But from what I have seen, I do not know what I'll do in that regard. Put simply, I am not certain that the CX-405 will produce better video than the R70 in normal well lighted conditions. If it does not, then the extra file size is a waste, and I would prefer to have the smaller file sizes of the R70's MP4 mode.

It looks to me that the only thing the CX-405 is storing in the extra large file size (compared to the Canon R70) is "image noise".

I will have to do more testing to tell for certain.

 

[rexbobcat]

What

 

[VidThreeNorth]

Summary with some expansion (so far):

This topic was going to be a simple report of what Sony's CX405's XAVC-S Codec was doing.

I started with a short history of the AVC and MP4 files in the two camcorders I have been using in recent years -- Canon R40 and Sony CX240. In those camcorders, the result was that Canon was generally better in outdoor daylight recording. Sony had generally better built-in sound, and their Back Illuminated sensor was better in lower light, including common brightly lighted indoor situations.

Recently I upgraded the Canon to the R70 and the Sony to CX405. The R70 was not a big upgrade. I needed to buy another unit "for safety", and any improvements were welcome. The Sony CX405 was bought because its advertising implied a substantial upgrade over the CX240, because of In Body Image Stabilization and the XAVC-S Codec. But Sony does not say what this particular version of XAVC-S was actually doing.

I did not state in my original post that I was considering replacing the Canon R70 with the CX405 for "most" of my 2D 1080P video recording, but that was my intention. If the CX405 was close enough to the R70, then I would rather use that one camcorder "most of the time". The only reason to use the R70 would be the rare situation where I would expect it do a significantly better job.

But the Sony CX405's XAVC-S codec does not have the YUV 4:2:2 colour sub-sampling that I wanted, and it is even doubtful that it has 10-bit colour (not totally disproven yet). Moreover, when I viewed the output of the XAVC-S recordings, they did look as good as the Canon R70 output. So far, I have not recorded truly comparable subject matter in the two camcorders, but "alarm bells" were ringing in the back of my mind.

More than that, the Sony XAVC-S output files are about twice the size the Canon R70. That means storing the Sony files will cost twice as much and take up twice the physical space.

This latter raises the following bottom line problem: If the Sony CX405 is going to cost me twice as much as the Canon for storage (and space is also a "cost"), then there must be a significant advantage to using it over the Canon. "As good as the Canon" is NOT good enough if using it is going to cost me more.

So that meant that I have more to look at than just the Codec. I need to look more generally at "image quality" and overall camera "performance" -- which I really did not want to bother with right now.


The second message was my attempt at isolating lens performance to see if that was causing problems. So far, as I concluded, the lens seems generally adequate. On further poking around, I think it might be a bit soft at the wide angle end, but then again, the Canon R70 lens is not perfect either.

So far, the Sony CX405 image seem to be a little less detailed than the Canon R70. Yet it uses twice the storage? This is not sounding so good.

The CX405 will replace my CX240, eventually. There is no doubt about that. But, so far, I can still see a good reason to prefer the R70 for outdoor, daylight recordings -- cheaper storage. If it turns out to have better image quality as well, then that just makes the decision stronger.

But I still have not recorded a truly comparable set of files to finalize this decision. So that is where I am now -- no real conclusion, but so far, it looks like I will (still) be using the Canon R70 for a significant amount of recording.

 

[VidThreeNorth]

Some Calculating:

I generally estimate my video file ratio (the amount I record over the final video) at around 5:1 - 10:1. So to produce a 5 min video, I expect to need 25 - 50 min. of recordings.

I took the data from a typical daylight video I recorded using the Sony CX405 recently:

"C0006.MP4"
2018-09-16 16:24:58
1,067,127,875 bytes
Length: 02:45
165 sec = 1,067,127,875 bytes
6,467,441.7 bytes per sec

1G = 1,073,741,824 bytes
1G = 166 sec. = 2 min 46.

A single layer DVD (for storage) holds a bit over 4GB. So that will be about 11 min 4 sec. So, just to store the camera files for a 5 min project is going to "cost" me around 2 - 4 DVDs? Keep in mind that this is still only "Full HD".

"01-Upld-Mod-17h34m45s073-C1.jpg"

This is a capture from the video clip that I took the data from. NOTE: Faces have been obscured, so if you are trying to judge the results, do NOT consider the faces.

A "C1" compressed JPEG of the same image BEFORE the faces were modified was 1,773,093 bytes.

 

[CherylL]

The two you mention, Sony CX405 and the Canon R70, are low end camcorders. If you are concerned about file quality then I would recommend one of the RX100 series. I have the iii & the iv. You may find these used or as an open box item. I think the RX100iii is selling around $300 used. I prefer the XAVC S over the AVCHD. They use different compression and are in different wrappers. You will have to factor in your processing power for editing HD video, adding effects, titles etc.

Hard drives are cheap these days so storage should not be an issue. The last DVD I burned was 3 years ago. I shot in HD and downconverted to widescreen SD. But, I exported the edited file in HD to give to the person on a USB drive too.

Video in general is a little flat so adding a touch of contrast and warmth gives some depth.

 

[VidThreeNorth]

RX100's are mainly 3x zoom with one at 8x zoom.

In the Sony CX405 and CX440, the sensor is 3.1mm (1/5.8") and the ZEISS Vario-Tessar lens is 30x optical zoom 1.9 - 57mm, F1.8 - F4.0. The 35mm equivalent view range is about 26.8 mm - 804.0 mm
In regular recording mode (not Low Light) minimum illumination is about 6 lux -- I assume that this is at the wide angle end with F1.8.

Take a look at this video recorded with the older Sony CX240:

"20160831 - CNE - Ricoh Coliseum - Toronto - Hit List Aerial-Ice Show v2"
""

This was all a single performance. I have no way of knowing how much of the zoom I actually used but I am fairly sure I used more than 20x. You can tell when I used the longest telephoto that the F-stop was getting near F4 because noise starts showing up.
[2018-09-26 23:13
I think I have to correct myself here. I have been watching my copy of this video a few times and I do not think I exceeded 20x zoom. On this camcorder 20x is about equal to a 600mm lens on a 35mm camera. The longest focal length in the video looks like around 400mm - 500mm. I doubt if I have ever really exceeded 20x on any of my camcorders yet in a "real" photographic situation. On a camcorder, even using a monopod, I would find it almost impossible to use.]

The Canon sensor is a bit bigger, but is not Back Illuminated, so it fuzzies out with noise in lighting like this.

All the exposures in this video are manual because theatrical lighting is always a mess. I kept it low for the colours. Actually, I have a better set of recordings this year, but I have not put it together yet. This year I was probably about 1/3 stop higher. Looking at this older clip now, I could have lightened this clip in post, and I am almost motivated to redo it again. Uh, no, I think I would rather move on.

The thing is, because Camcorder sales have dropped due to cellphones, there has been no development until you get around $1,000 US range. So there is this big jump and what I can do with a $200 - $300 camcorder is as good as what is out there up till you are around $800, which gets a 4K camera with maybe 10 - 20X zoom. And if you think that an interchangeable lens system can even duplicate this, try, maybe $5,000, used, if you are very lucky? A 30X F2 range variable speed POWER zoom with good working image stabilization?

I have had a chance to try the sensor shift image stabilization in the CX405 a bit and, uh, actually, the digital stabilization in the CX240 might have been better. Again, it's a bit early to make "final conclusions", but at the far Tele end of the zoom in "Active Stabilization" mode, the sensor shift seems
to add jitter. Then again, I was expecting to use "Standard Stabilization" anyway, because it limits the zoom to optical.

[added @ 13:02]
I was looking at the new clips and comparing them with this older video and I remembered that I was recording everything back then with a polarizer. Two differences were that the ice glare is reduced, which brings out the skaters more, and of course, it reduces brightness by about ND 2.5, which is a little more than 1 F-stop. So some of the image noise in the telephoto would not be there if the camera was used "normally". But even with the polarizer the CX240 was out-performing the Canon R40 in tests. So the Sony sensor is really good in that regard.

And yes, I find I do prefer the ice show clips with the polarizer on, but it is not "terrible" if I do not use it.