@Frame: I don't know/use Premiere, so I cannot recommend any details on it. But generally speaking, the idea is not to avoid the first/last frames when "importing" the clips (assuming that "import" here means: telling the software about the available .mts files - without transcoding them into some different format), but to set the "in" and "out" points of your clips (afterwards, e.g. when putting them into the timeline) such that their first and last 12 frames won't be used as a source for your final output rendering.

You mention that the part you uploaded was 30s into a larger file: Yes, and the excerpt starts with data for an I-frame, followed by data for b-frames that have a Presentation Time Stamp (PTS) that is before the PTS of the I-frame (because the file needs to contain frames in an order that allows decoding of P- and B- frames based on differences to frames that were decoded before), but the second I-frame that is needed to interpolate the 2 B-frames is not part of the file (it would have been in the part before the excerpt). So the decoder sees two B-frames that it has only one reference frame for, and at that point it may decide to discard or partially decode those frames. ffmpeg seems to opt in favor of the partial decoding under certain circumstances: when there already was at least one other decoded frame. This would explain why ffmpeg does not decode the B-frames with missing reference when cold-starting a replay, but it does decode those B-frames both when looping (using the last decoded frame from the sequence instead of the missing reference) and probably also when "seeking" in a file - e.g. when the NLE software tries to single-step to a frame where to start, but has already decoded others in the process of getting to the right position.

The reason why I see that effect often is that my workflow starts by cutting relevant parts from the .MTS files I recorded in the compressed domain (without re-encoding). I need to do this, as I often take lots of long shots (the "actors" being animals not allowing me to request their action on command), and if I would not decimate the material quickly after shooting it, it would use up excessive amounts of harddisk space. So I live with the mentioned work-around until I can find a better solution, which could be:

• fixing avidemux26 to get rid of the incomplete B-frames when cutting in the compressed domain
• changing kdenlive/mlt/ffmpeg to do the "seeking to the right position" in a different way, avoiding to produce initial garbled frames.

@rapsoe for logging its great - rain or no - trust me even in broadcast progs - the chaps n chapesses that log and capture any ingest ... rain? Had a QC dept this week pass an audio mix (that I hadn't even laid back until the day after i.e. the all the way to 0 offline, crunchy drunk disco dancer music edit mix) on a well (multi million watched) series in it's nth series - joy. Ravensbourne where art thou with your brown cords and flares lol

@LPowell Looking forward! Thanks for all your great work!

@luxis The stability revisions I'm testing are intended to insure failsafe operation on all Class 10 cards, regardless of their rated maximum bitrates. In my experience, the advantage of the 95MB/sec Class 10 cards are primarily in their ability to support 4GB file-spanning at 100Mbps. Although standard 30MB/sec Class 10 cards are capable of reliably recording individual 4GB files at 100Mbps, they usually fail to span. With Flow Motion v2, you may see a brief write-speed error message when file-spanning fails on slower Class 10 cards, but the initial 4GB MTS file will be saved intact. FM2's 60Mbps 24L, FH, and H modes should span on standard 30MB/sec Class 10 cards, as well as higher rated SD cards.

@towi well, it was not the solution, apologies. My GH2 worked without problems for one day shooting HBR, but I emptied my card, format, and it crashed again... Shooting test charts with HBR is strange too : usually it crashes after a cold start, but if I pull out the battery and shoot again, no more crash. Thanks for sharing your settings, I'll give a try :-)

@Lpowell : great news! And thanks for your amazing work, FM 2.0 makes the GH2 shine! 24p is simply amazing. Panning footages are so fluid and movements so smooth that it's hard to belive it's shot with the GH2 :-)

GH2 Flow Motion v2.01 Update

This update addresses reliability issues with HBR, FSH, and SH video modes in both NTSC and PAL. It has been successfully tested for failsafe operation under all shooting conditions and options, including 80% Slow Motion and ETC zoom modes. The stability adjustments in Flow Motion v2.01 should produce little or no visible reduction in motion picture quality in any of the updated video modes.

In response to numerous requests, I've enabled and verified reliable operation of Flow Motion v2.01 with extended ISO levels. By default, all levels up through ISO 12500 may now be selected for use with any NTSC or PAL video mode. In practice, however, levels above ISO 3200 produce considerable noise, which may require excessive bitrate and perceptibly reduce motion picture quality compared to lower ISO levels.

Download the Flow Motion v2.01 Update at the following link:

http://www.personal-view.com./talks/discussion/3337/gh2-flow-motion-v2.01-update-failsafe-hbr-720p-modes/p1

@LPowell Thanks for the update ! I will try this with my 64g/95 card today. I will post later with results.

Fantastic! Thank you for this quick update. I will try it ASPA! Your samples are impressive. With FM2, I've found that an HBR footage re-encapsulated as progressive has virtually the same image quality than a 24p footage.

@LPowell. I just tested FM2.01 in 720p60 SH mode with Sandisk Extreme 32GB 30mb/s and it failed with write errors.

Tested FM2.01 in 720p60 SH mode with Sandisk Extreme 65GB 95mb/s...all good now! even etc mode is fine. 14-140 kit lens Iso 320 ....no card write errors this time..Yea!

Thanks @LPowell !

@Luxis. I use two SD class 10 cards, Genuine Sandisk Extreme HD 32GB 30mb/s and Genuine Transcend SDXC 64GB. I only use Panasonic lenses for GH2, 14-140mm, pancake and 100-300mm. The scene is nature (trees, flowers in the sunlight). I tried Intelligent Auto (IA) mode and Nature setting that comes with camera and sets the various values for you. I think one of the factors effecting is the Panasonic lenses that do continuous zoom in AFC mode and OIS on. 720p H mode works fine. It is the SH mode that fails. I have found out the the limit before it fails with write error is 55 MB/s. Once details get higher and it gets close to 60MB, it fails. The way I have gotten around it in FM2.0 is to reduce the bitrate for 720p/1080 from 99/55 to 77/55 along with associated top and bottom bitrate settings. This modification keeps the SH mode to approximately 55MB therefore avoiding the write errors. I have read that others have no issues using other lenses but Panasonic lenses seem to put more stress on the sensor.

@Zaven13 Thanks for your test report. I'll confess that I've never actually tried shooting video in Intelligent Auto mode before. As you found, Flow Motion v2.01 can fail under Intelligent Auto when used in SH mode. However, the patch does appear to work reliably under Intelligent Auto in 720p H mode, and produces bitrates comparable to your personal modifications, which have the drawback of degrading the bitrates of 1080p HBR and FSH modes as well. For users who find Intelligent Auto useful for 720p videos, I'd recommend the workaround of using 720p H mode instead of SH mode for this purpose.

@LPowell what did you exactly change from FM2.0 to FM2.01?

As I want to use it with no Panasonic lens (Olympus 4/3) and with no AF - which both seem to put more load on the codec/cpu. Card is Sandisk 30MB/s. Will FM2.0 work or should I use 2.01? Is the picture quality lower with FM2.01?

And thanks for your great work!