Tragic Lantern for 6D

[noisyboy]

Maybe sticky half shutter is on?

Turns out that I had trap focus set to continuous which was the cause of the problem. Thanks

So with resizing, I've kind of been doing that at the moment but I wondered if anyone had mathematically worked out what percentage it needs to be horizontally transformed by within a 1280 wide comp to get a "correct" image. I'll have a play and share my results.

[noisyboy]

Okay - so now I AM impressed!

Just been playing with the zoom function with DNG Burst and man - the image is amazing! Pointing at a load of trees I see no noticeable aliasing at all and post sharpening seems to work amazing on it!

[1%]

If you don't see the pics when connected VIA usb then you have to use a card reader.


I've just been doing it by eye until it doesn't look stretched.

[a1ex]

The images from 720p should be 16:9, so just enter that ratio in your image editor.

[1%]

How do I find the location for CBR handler... I can't find the register interrupt function for edmac.. for write edmac I found some stuff but it only gave me FFXXXX and not names.

[noisyboy]

The images from 720p should be 16:9, so just enter that ratio in your image editor.

Sweet - so I guess just create a 1280x720 or 1920x1080 comp and just "fit to comp". Great - cheers a1ex and 1%. Again 

[fotojohni]

I don't know what improvements have been made in this area, but I want to do an update on the bitrates.  IPB is not safe to use with ANY bit rate modification.  It will drop frames when put under pressure, it happens very rarely, but I was photographing a tree on a windy day and everytime the wind blow and started rustling the leaves the read out would tickle 80%.  I looked back at the footage and sure enough, it had dropped frames. 

On All-i with the same settings it never worked for more than a second.  So I turned it up a tiny bit to qp8 and it was pretty good, same sized footage though, which is a large ~1GB/min. 

[1%]

Yea, you're recording fixed QP so it will drop frames when it can't save them at QP whatever. Thats why I increased the max qp in the configs and don't do fixed QP in general.

[fotojohni]

Yea, you're recording fixed QP so it will drop frames when it can't save them at QP whatever. Thats why I increased the max qp in the configs and don't do fixed QP in general.

Right, I just recorded two files, one minute long on RC at bitrate = 60 Mb/s.  Calculated out that's exactly 450 MB for one minute.  I looked at the files and sure enough, exactly 468MB and 471MB.  What would be nice is if there was an RC where you set the target bitrate at something like 30Mb/s but have a max of something like 120Mb/s.  I thought that's what VBR was doing because of init QP, I assumed it just aimed for the init QP then altered if it went over the max bitrate.  I can imagine that kind of algorithm is ridiculously hard to implement on the fly though, and probably takes a few passes to get right, which is why arri, red, and BMC guys just say eff it and record the raw.

It's nice to be able to fall back on RC though if you just can't get the VBR to a nice number without it blowing up.

[1%]

Run vbr without any overrides... try 30P if it freezes in 24

The algos are all in the encoder part of the digic, I'd have to find where to manually set current QP, I don't know if you can just do it on the fly... that location is "init" or start qp and is probably only read once.

CBR config is for fixed QP/override like you're doing. If you look in the IVA params thread on 5d3 some "max" rates were found before it starts overloading the endoder/writing. That point for CABAC/CAVLC is different and ALLI/IPB/IBB too.


The way this works is QP is quality, so QP 1 can be 30mb/s to 400mb/s depending on your scene. Running fixed QP with VBR gives you the extra profile stuff and CABAC but when you look at the files, every frame will be at that QP you set. (Why I kept saying don't set initqp with configs) When the frame becomes like 2MB and the buffer gets hit with 24 of them, if the writing can't keep up the encoder drops frames. It can't really do anything else, it can't lower QP, you told it not to.

The RC will go up to 100MB/s and over if your scene is complex enough. Some scenes are just 60mb/s, they are static with not much going on. When you look in the file, QP is 1 already. 

Tested it before flush rates and evened it out. The old one had a limited range of QPs and then when scenes got big and it ran out it did something similar (truncated frames). You can tweak the min/max to give you higher rates but I kept it 1-30 so it wouldn't run out, esp with upped flush rate.

BTW it is possible to get a clean file at ~300-400Mb/s fixed QP1, just its very hit or miss.

[teo770]

Funny: i quit the GH2, tired of looking for "the ultimate hack" to have a good compromise btw image quality and stability, telling myself, with new Canon's high bitrate encoding.... it can only be better out of the box and..... bitrate hacks are coming to ML too !

External encoders like the Atomos Ninja records at 220MB/s (4.2.2)
Canon 6D and 5D Mk III share the same bitrate of 90MB/s (4.2.0), no ?

How far could be pushed the bitrate on those cameras ? twice 90MB/s ?
With what kind of stability ? Would depends of memory card speed too, of course.

What's the "reasonable" goal ?

[1%]

Reasonable is to just use RC and get 60-120Mb/s depending on scene. The SD card writes up to 50MB/s only.

[fotojohni]

The algos are all in the encoder part of the digic, I'd have to find where to manually set current QP, I don't know if you can just do it on the fly... that location is "init" or start qp and is probably only read once.

well if you do find it this is what the algo could look like:

if (currBitRate > maxBitRate){
...currQP = currQP +1 //try to get below max bitrate
...//alternatively: currQP = currQP + (currBitRate - maxBitRate)/K  (just to really cut it down if it needs to)
}

else if (currBitRate < maxBitRate & currQP > initQP){
...currQP = currQP - 1 //try to get to target qp
}

else{
...//it's all good (or maybe there could be a statement for a min bitRate if you wanted and it would bump the QP for that)
}

I'm not trying to be condescending or whatever (if this looks crazy just call me a moron and move on), I only understand the concept barely myself, and I have no idea how QBasic or whatever the firmware language is works.  I just think that this is a good way to do it because generally, the camera gradually comes into a complex scene, when you pan or focus pull into it, it usually doesn't just happen and if the current bitrate was signifigantly higher than the max then it would need to really cut the QP fast, so there could be a scaling factor for that.  It might be a good idea to start the currentQP signifigantly higher, like around 40 higher, than the target and then work it's way down, it would take not more than 2 seconds at 24p.  I don't know, just a thought.  I can't do any tests right now because I lent all of my cards and 600D out to a friend, I'll do more tests when I get those back.

[1%]

That stuff is inside the encoder. Don't have that kind of control yet. If I find something like slice in this encoder I can do that.

[noisyboy]

Cheers for the bin dude. What do I have to do to test this baby out? Got it all loaded up but having a wee bit of bother finding the hidden treasure.

[noisyboy]

Ps. Please feel free to tell me to sling it if it isn't ready yet - I know you don't need the hassle of being tech support to the n00b when you're trying to get this done. I'm cool with that.

[noisyboy]

In fact I have it! Thanks again!

[1%]

Load modules... its on the bottom of the movie stuff menu.

[coutts]

shot 8gb of 720p 24fps sample footage, working on processing this now I'll post it up when it's done. Just some night shots and random footage at a store.

[1%]

One of the 960s is pretty much continuous. Sucks we're stuck with this crappy card speed.

[noisyboy]

Works great (as does the raw2dng extractor) so congrats

I take it the number of frames I'm achieving (59 at 30fps on first tests) is to do with the limitations of my cards read/write speed? Sorry if I misunderstand that.

Also, are there "optimum settings" for achieving longer record times as with the Silent Burst method?

I also noticed that the default frame skipping is set to 1. Does 1 mean it isn't skipping any or that it's skipping every other 1?

Finally, if I disable RAW and shoot YUV am I likely to achieve similar record times or possibly longer?

Again - this is incredible. To the many budget film makers like myself out there - you are all officially superheros!

Ps. It stopped recording automatically so didn't encounter any freezing.

[noisyboy]

One of the 960s is pretty much continuous. Sucks we're stuck with this crappy card speed.

I posted just after you did so sorry I missed this as it answers one of my questions

[1%]

LV_REC and Raw_REC are different... you have both. Raw rec only has settings for W/H

[noisyboy]

Ah! Super cool Cheers!

[noisyboy]

1320x540 seems to give me the highest continuous recording capability without interruption. And oddly, I get better performance with 1320x720 than I do with 1280x720. In fact I get about 17 seconds @ 25fps with 1280x720 and about 30 secs before I start getting the odd frame skip with 1320x720. This is pretty impressive!