GH1 firmware research volunteers required

you're not suffering dropped frames... in one GOP you managed to get 8 P frames instead of 7 :eek:
keep in mind: no dropped frames & take a look at the following screenshot (the 1080 clip at the top, the 720 clip at the buttom)... the gaps in the 1080 clip are not missing frames but virtually nonexistent (regarding framesizes).

Hmm, thanks for the analysis, but I'm afraid I don't quite get what you are trying to say! How could this happen?


Also, regarding the NTSC patch:
It seems to work. The following is reported for 1080p file:

Bit rate : 13.4 Mbps
Width : 1 920 pixels
Height : 1 080 pixels
Display aspect ratio : 16:9
Frame rate : 29.970 fps

And this for 720p:

Width : 1 280 pixels
Height : 720 pixels
Display aspect ratio : 16:9
Frame rate : 59.940 fps

However, somehow I get a "Recording Cancelled due to card speed writing limitation" error when I try to record in 1080p, longer then 10 seconds. However, in this version I checked ONLY the NTSC conversion and the version change, nothing else! How could this still be the case? Maybe the card is busted :(
 
Yeah, try other card.
And 29.97fps is strange as you must see 23.97fps (but in 29.97*2 wrapper).
Try using wrapper 60i patch also.

And believe that our collector could make Western Union transfer (or PayPal if Chibs prefer to get money using it) to buy him really fast Class 10 card. I think we could provide us a quote for this card in his country.

Also I hope that some of NTSC guys could test PAL switch.
 
It'll be good to see someone with good AVCHD encoder implementation knowledge.

that would be great but as long as we only mess with variables & don't stand a chance to change something elemental (i.e. codec implementation) it won't help us that much...

Hmm, thanks for the analysis, but I'm afraid I don't quite get what you are trying to say! How could this happen?

take a look at the txt file. the recording starts & the first 63 frames are incredibly small. thats what's bothering, cause there is no bottleneck till then, buffer ist empty & frames so small that write speed is no issue. then, the recording starts for real & you get the pattern: # of big frames
(both I/P frames about the same size) followed by # of small P frames

30 big - 7 small frames - 19 big - 5 small - 17 big - 7 small - 19 big - 5 small - 17 big - 7 small - 17 big - 8 small - 16 big - 7 small - 1 big - 7 small - 17 big - 7 small - 18 big - 6 small - 16 big

whats causing the small frames intervals in between? sensor buffer full? card to slow? DSP? codec not able to keep up the pace with that data rate?
 
Tester13 - keep this up man. Things are looking promising here. Hopefully this will force the hand of Panasonic. If it doesn't then I am not sure what will..........As I have siad already and not sure if this can be done easily but HDMI out would make this thing a giant killer. Will test this out tonight on PAL GH1.

Just sent $50 your way guys - nice work and keep it up.
Henry
 
Last edited:
Is anyone testing this with an NTSC GH1? The mud fix is looking promising on PAL, just want to see if it holds up on the NTSC end too. And of course faster cards, if it's possible to try Class 10 (Not the cheapo kind)with the higher bitrate settings and report back that would be fantastic.

Great work, keep it up. :thumbsup:
 
that would be great but as long as we only mess with variables & don't stand a chance to change something elemental (i.e. codec implementation) it won't help us that much...

I believe that most parts are implemented in hardware. But later we can completely rewrite whole functions in AVCHD codec, if we could reverse them.
They are referenced by offset table, so we could use firmware empty space and change one offset, and here we go.
But we are far from it.

If you look at ML they did the following - loaded SIG file to find most OS functions and PTP functions, found screen buffer address and information overlay address almost instantly and made modifications for this parts.
AGC in ML is modified using direct IO ports work (as 5D firmware have multiple strings and it is not hard audio related things).
Our hope for AGC is that GF1 allow us to know something here as it have audio level related strings.

whats causing the small frames intervals in between? sensor buffer full? card to slow? DSP? codec not able to keep up the pace with that data rate?

Could you explain it in simpler terms? For plain dummies like us :-)
How long GOP works relating to this?
 
Is anyone testing this with an NTSC GH1? The mud fix is looking promising on PAL, just want to see if it holds up on the NTSC end too. And of course faster cards, if it's possible to try Class 10 (Not the cheapo kind)with the higher bitrate settings and report back that would be fantastic.

Chibs - try to find good fast Class 10 card in your vicinity. Report it's price back.

Car3o - send him WU payment (it is instant thing) for this card, as it looks that it is necessary to understand encoder restrictions and hardware restrictions.
 
I did that test (check out the thread) and I found the image looked unacceptable to me....I recall you disagreed.

It just looked like 30p with a frames dropped because dropping every 5th frame does not change the incremental difference between the other frames.

i posted 2 new clips on that thread deleting the 5th frame, vs the 4 or 7th frame and nobody ever chimed back in. however it's not really important anymore, what is is if it can be done and then tested and whether or not it's acceptable for everyone else and how it looks.
 
some quick math based on Chibs stats: his trial with 1080 had avg frames about 85k, 320 frames, resulting in 26MB for an 13s clip. thats 2MB/s (aka class 2). lets change this a bit. assume all his frames would have 144k, would result in 44MB, thats ~3,5MB/s (below class4 :eek:)
there must be some other bottleneck & i fear its in silicon. Chibs: do you have some other class 4 card to test with?
 
Without having to read through all of the thread - is there included in the patch the option to make the Pal MJPEG turn to 25p?I seem to be missing this - If this is answered somewhere please tell me where.
 
Even though this might be the most ingenious and elaborate Russian hoax to get a free GH1 and class 10 SDHCs, I love this thread! It is the only thread I check lately.

AGC! AGC! AGC!

Keep up the great work guys!
 
Without having to read through all of the thread - is there included in the patch the option to make the Pal MJPEG turn to 25p?I seem to be missing this - If this is answered somewhere please tell me where.

No we don't have it jet.
Only 24p and it is just skipping frames, according to testers.
We change MJPEG encoder settings, and not real sensor framerate.
 
Even though this might be the most ingenious and elaborate Russian hoax to get a free GH1 and class 10 SDHCs, I love this thread!

Yeah :-)
In this case I only do not understand why I made GX20 hack efforts and run LG TV's internals research forum. May be to get cargo container full of gold?
 
Yes, I do have paypal. I have sent you a message, Car3o. The cheapest I could find that I could get tomorrow is 38 euro for a 4Gb Sandisk Extrem III Class 10 card. Sandisk should be reliable in their speeds, I hope?
 
some quick math based on Chibs stats: his trial with 1080 had avg frames about 85k, 320 frames, resulting in 26MB for an 13s clip. thats 2MB/s (aka class 2). lets change this a bit. assume all his frames would have 144k, would result in 44MB, thats ~3,5MB/s (below class4 :eek:)
there must be some other bottleneck & i fear its in silicon. Chibs: do you have some other class 4 card to test with?
I have one class 4 card, a Kingston. I will go look for it now to test this, thanks for the math!
 
I believe that most parts are implemented in hardware.

that's my guess too

Could you explain it in simpler terms? For plain dummies like us :-)
How long GOP works relating to this?

the plain dummie is rather on my side of the internet ;)
i don't know why it fails. based on the numbers it almost makes no difference between I/P frames (they are about the same size). after some time it seems like some buffer (either readbuffer from sensor or some write buffer?) gets full & frames (only P frames till the next I frame) are very small with no infos. as long as we dont know how the GH1 handles GOP & how the frame sizes get allocated its stumbling in the dark.

@Chibs: can you test 1080/720 with GOP size 1 (means only I frames). even with class4 this should be ok, i'd like to see what the GH1 does in this situation.

i'd like to be proven wrong about the card speed (that its irrelevant when > class4)...
 
SD Class X is very strange thing.
It could be anything. By idea 4 means 4MB/s and 10 - 10MB/s.
But reality can be different.
Still do not understand why they moved from CompactFlash with clear concept having complete controller on card side.
 
Back
Top