Results of my first XDR Timelapse

[Mark Job]

Originally Posted by Aaron Newsome

This confuses me since I-frame only codec has no GOP. It's intraframe, meaning each frame stands alone, not as part of a group of pictures like long GOP.

Maybe it would make more sense if you just wrote each image to a single frame Quicktime / MXF.

....Hi Aaron:
Yes. I find this very confusing, but much to my surprise, I actually understood Mike's explanation. I don't understand exactly why CD has chosen to impliment MPEG I-Frame capture within a Long GOP format, but doing so has taken on the limitations of the Long GOP format indeed. Their file size ratio winds up being way too small to be practical.

......I also think creating an image sequence is another way to record digital frame grabs. All NLE's will accept image sequences in JEPEG or Targa usually. This is how we will do Time Lapse with our SSDR.

[Dan Keaton]

Originally Posted by Aaron Newsome

This confuses me since I-frame only codec has no GOP. It's intraframe, meaning each frame stands alone, not as part of a group of pictures like long GOP.

Maybe it would make more sense if you just wrote each image to a single frame Quicktime / MXF.

Dear Aaron,

Our I-Frame Only, is exactly what is says, it is only I-Frames.

From a purely technical point of view, we create (for 1080) 15 I-Frames in a Long-GOP sequence. For 720 it is 12 I-Frames in a Long-GOP sequence.

So, they are all I-Frames, so it is intra-frame recording.

In nornal operation, this has no side effects whatsoever. At the end of a recording, we complete the GOP before we stop recording. This takes up to 1/2 second, no big deal.

In timelapse, we want to finish the Long Group of Pictures before we close the file.

In the most extreme case, if we were recording 1 frame a day (not a normal practice), it could take up to 14 days to close the file. If we were recording at 1 frame a second, which is the most common case, it would take up to 15 seconds to close the file.

So, for a entire timelapse sequence, the last GOP, in the last file, will be partially padded out at normal speed. This allows us to close the file prompty.

[Aaron Newsome]

What I mean Dan is, don't most other I-Frame codecs have GOP = 1. I mean, if each frame stands alone, isn't a long GOP redundant?

[Dan Keaton]

Dear Aaron,

Other systems could easily have 1 frame in a GOP, I do not know.

We have not investigated having 1 frame in a GOP.

1. I do not know if we can do this.

2. I do not know if we do this, if any editor would support it.

In other words, we have not considered setting the GOP to 1.


We expect people, when using the timelapse feature to record slightly longer, or a lot longer than what they will actually use. Thus, it should not be a big deal to cut off the last 15 frames of a multiple hour timelapse sequence, especially when these last 15 frames represent 1 second.

[Mark Job]

Hi Dan:
Yeah, this concept bends the mind. Why use a Long GOP MPEG video method to Record I- Frame MPEG video ? Surely there must be a very good reason for this ? Would it not be possible to simply record Time Lapse via pure I- Frame MPEG Video File Structure ? In this way, using a purely I-Frame MPEG video file structure would end the need to have a speed jump at the end of a time lapse sequence to accelerate the file closure operation would it not ? Or is there a technical consideration here, which strictly enforces the requirement to capture the I-Frame video in a Long GOP format ?

[Dan Keaton]

Dear Mark,

After recording a long timelapse sequence, in post, cut off the last second, of the last file that we created and you will be ok.

All of the other sub-clips (files) will not have the speed bump. Only the very last of any timelapse sequence.

[Mark Job]

Hi Dan:
There is a fatal flaw in what you are suggesting. What you propose is viable *only if you successfully take the TL sequence using One (1) recording session.* If you have to shut down and restart for any technical reason (Like changing a battery) then this speed jump at the end of each sequence would ruin the flow of the sequence when you have to join two sessions together to make one sequence. Think about it. Any speed jump would have to be edited out, of course, and as soon as you do that, then you have a big missing piece of your timelapse progression, therefore you cannot match multiple sessions together properly.

Caveate: You must take your timelapse sequence in *One* (1) recording session. Bring big capacity battery (s) to shoot.

[John Mitchell]

Guys there is a fundamental misunderstanding going on here. MPEG is a long GOP format - an I frame only format merely fills the long GOPs with I frames. A 1 frame GOP would "break" the MPEG format rules for the codecs involved - it's a standard and your NLE expects it. There is a standard technical way to get around this. You can "fool" the GOP into thinking it is full by using a flag that tells it to repeat the one frame as many times as necessary for 1080 or 720 - and the good news, it takes up no extra space. This would at least allow recording to stop instantaenously.

I don't know if you can then force the GOP to play back as one frame only - that would involve another flag or specialised software. It then becomes non-generic or needs post processing. If you have a 35mm DSLR that will do timelapse it currently seems like a better option than the XDR or Nano, but if you don't then the XDR seems like a reasonable, if imperfect solution.

[John Mitchell]

Originally Posted by Mark Job

Hi Dan:
There is a fatal flaw in what you are suggesting. What you propose is viable *only if you successfully take the TL sequence using One (1) recording session.* If you have to shut down and restart for any technical reason (Like changing a battery) then this speed jump at the end of each sequence would ruin the flow of the sequence when you have to join two sessions together to make one sequence. Think about it. Any speed jump would have to be edited out, of course, and as soon as you do that, then you have a big missing piece of your timelapse progression, therefore you cannot match multiple sessions together properly.

Caveate: You must take your timelapse sequence in *One* (1) recording session. Bring big capacity battery (s) to shoot.

Mark - I don't get this - if you had to stop recording for any reason you would ruin the flow anyway unless you could time the restart perfectly? Have I misunderstood you?

[Mark Job]

Hi John:
Actually, you make an even more fundamentally valid point, which is what I was trying to express but in different words. In case you want to *salvage* a 5 hour sit down because a battery has managed to last allot less than you thought it should, then under normal conditions there cannot be any hope in doing so if you have any kind of *speed jump* at the end of the recording session you just had to close before you battery died completely. If you cut out the speed jump and join it onto the next session, then you will have in most cases a 'missing piece' of TL progression - checkmate ! If you leave the speedbump in, then you also have a nasty interruption in the timelapse flow - checkmate once again.

You have two possible solutions:

A) You close one session and open another within the interval you are shooting. (Can this be done ?) Maybe ? What about the speed jump you now have in the middle of your time laspe scene ? You could.......*Possibly* cut out the jump and frame blend in post and hope no one notices. This sometimes can be done. I have actually accomplished this once.

B) You capture everything in one session and then record longer than you need and let the sequence speed jump away at the end and simply remove the jump part in post. (Dan expressed this idea)

* I still think this speed jumpy thingy way is a fatal flaw, but what to do if it takes another 9 hours to close your session without one ??? Hmmmmm ? This drives me back to my earlier inquiries as to why do we have to capture TL's using an I-Frame within Long GOP method ? Can we not use straight I-Frame MPEG recording structure ??

I think you have pretty much failed capturing your TL sequence if you don't get it all in one recording session, so bring large capacity batteries to the shoot.

[Alister Chapman]

To be honest if you are doing timelapse sequences with intervals of a minute or more the NanoFlash is possibly not the most appropriate tool anyway. You would be better off with a stills camera for shoots that are going to take hours or days. You don't need massive batteries, you can shoot at very high resolution and tying up a $1000 DSLR for hours, days or weeks makes more sense than using an expensive HD camera and NanoFlash. IMHO Where the NanoFlash will be most useful is for simple things like cloud and skyscapes, traffic and other moderate speed-ups.

[Gints Klimanis]

While your DSLR suggestion is indeed practical, it would still be useful for the Nanoflash to save uncompressed image files for time-lapse if that is possible.

Sometimes I am asked to produce images from shoots or have to use a zoomed/cropped frame in a video. Yes, a DSLR is a better tool, but I'd rather not carry and fiddle with one while I'm busy doing video.

I'd like to have a Nanoflash feature to save an uncompressed image file on command. It takes more time to search footage and grab a frame. And then, it is already compressed.
On a 3 chip camcorder taking 1920x1080p, we have the equivalent of something like a 4-8 MPixel Bayer sensor DSLR.

There is another application for frame bursts. DSLRs have peaked at 11 fps (D3 in crop mode). Imagine a device that would record a burst of 30 or 60 frames, then save them as images for you to choose. If only there were the equivalent of a DSLR RAW mode in which unprocessed sensor data were transmitted via HD-SDI.

[Mark Job]

"If only there were the equivalent of a DSLR RAW mode in which unprocessed sensor data were transmitted via HD-SDI."

....Hi Gints:
There is. It's called the Canon XL H1 with the HD-SDI output. This is the full uncompressed RAW signal of 1920 x 1080 straight off the 3 x 1/3rd inch sensors.

[Aaron Newsome]

That's incorrect Mark. The Canon camera outputs a very processed signal via the HD-SDI. First, half the color is thrown away which makes a 4:4:4 signal into a 4:2:2 signal. It's no where near straight off the sensor. It has white balance, gamma curve, color matrix, knee, and various other processing of the signal from the sensor.

The Viper on the other hand, does deliver raw sensor data in Filmstream mode, completely unprocessed, only fitted in 10bit log 4:4:4 :-)

[Gints Klimanis]

While a push button for an unprocessed (or even a processed full resolution) frame grab may seem like a consumer-gadget function, I'd really like to see someone cover the high speed, higher resolution 24-60 fps burst photography space at a lower cost.