@LPowell , can FH quality mode be modified for lower bitrates (1080i 50) ? , I would like to get about 2h 15m on 32GB card ? If it can be done without interrupting other settings please let me know where can i change them ? Same for 720p mode the H ?

@driftwood Thanks for the suggestion. I uploaded a comparison between Cluster v4 and Flow Motion v2 in 1080p 24H mode in the first post on this page.

@LongJohnSilver The left edge of the frame looks like it's illuminated in bright sunlight, which narrows its contrast. That makes it look less detailed that the darker areas of water. At the distance this video was shot, the water surface ripples are tiny, making them look more like random noise than genuine image details. Perhaps this scene would look more cinematic with a diffusion filter?

@LongJohnSilver Your second video is not as intensely detailed as the first, and FM2 recorded it in its standard 100Mbps mode. Both Stream Parser and Elecard StreamEye Slim reports look flawless, indicating that the encoder was operating at optimal quality levels as intended. (The target QP level for Flow Motion v2 is 17 for both I-frames and B-frames.) With an all-Intra patch, you'd need a bitrate of over 175Mbps to maintain comparable encoding quality in each of its I-frames.

@LongJohnSilver @LPowell

From what I see on vimeo, its not only the water in the bottom of the frame, but also the distant water in the middle. There is rain/insects hitting the surface randomly producing small circular waves. If you look close enough you can see that the footage jumps in time in that area (about 1/2 second).

(I hope its not the vimeo compression/playback thats fooling me here;-)

Following up on Driftwood's recommendation below, I shot a test comparison of his Cluster v4 patch with Flow Motion v2. For this shot I used a handheld Leicasonic 14-50mm f2.8-3.5 OIS lens in 1080p 24H mode. Here are links to each video, which include downloadable original MTS files:

Cluster v4 in 1080p 24H mode:

Flow Motion v2 in 1080p 24H mode:

Cluster v4 is indeed a high-quality long-GOP patch, with peak bitrates up to 100Mbps. However, its P and B-frames are encoded with average quantization levels that are coarser than the B-frames in Flow Motion v2.

Using Elecard StreamEye to examine the original MTS files, average QP level of the macroblocks in each Cluster B-frame is about 28. If we combine this factor with the average of the first four entries in Cluster's B-frame Quantization Table, we can calculate a good estimate of the average quantization quality used to encode the frame. For Cluster v4, this averaged table value is 8 and combining it with an average QP level of 28 produces an effective quantizer of about 8.1 (lower values produce finer quantization quality).

In the Flow Motion v2 case, the average QP level of macroblocks in each B-frame is 20. The average of the first four entries in the FM2 B-frame Quantization Table is 11. Combining these factors produces an effective quantizer of about 4.3. These figures indicate that Flow Motion v2 B-frames are encoded at a quantization quality nearly twice as fine as Cluster v4 B-frames. OTOH, the FM2 MTS file is nearly twice as large as the Cluster v4 file.

Here is an unfiltered frame grab from the above videos displayed at 200% magnification. The Cluster v4 frame at the top looks very good, with no visible macroblock artifacts. However, its limited bitrate was not able to capture quite as much flowing water detail as the Flow Motion v2 frame below it.

I've always wondered whether this water rendering problem is inherent to water or because water is nearly always toward the bottom of the frame. Sometimes the encoder mis-estimates the base Q value to encode busy scenes. What happens is that there ends up being a lot of "skipped" macroblocks toward the bottom of frames because the encoder has run out of bandwidth.

An interesting test would be to record the same subject with the camera upside down. This would tell us whether the codec is failing because the scene is too busy, or whether it fails because water is typically lower contrast than grass or trees and the codec "sacrifices" lower contrast detail in favor of higher contrast detail.

@mo7ies

I don't know I was there for real shooting not for testing. It's the second time I fail...

Hi Lee,

thank you for checking my shot. I see your point. On your FM samples the river bottom was of nude rocks and the water was flowing much faster.

Are you shure it's a problem of bitrate? On stream parser I see 119 Mbps (I'm a noob) but I uploaded another shot where the water is only in the lower partof the frame (about 88Mbps) and still the gentle water circles are rendered in a artificial way. Could it be a weak point of the encoder? Does it mean that with the current GH2 these kind of shots cannot be made? Am I the only one filming nature subjects like these?

The problem shooting in 24p above water and 30p underwater is mixing the clips on the timeline...

I would be curious to see an unhacked GH2 on this subject or another canon camera :)

Thanks a lot

PS

After the bush of death now we have the lake of death :)

Here it is: the water is still frying....

I'll retest with FM2 this week. I've been meaning to give it a go anyway :)

@jrd I haven't tested with FlowMotion2 but when I did my last round of tests back in February I found that I had considerably less timeline responsiveness with FlowMotion compared to Cake and Sanity (using PPro 5.5). It was enough on a problem that I went with Cake for the event I was filming.

I didn't have time to fully diagnose the problem and I haven't tried with FM2. YMMV. I didn't report this at the time because I didn't feel I'd investigated it enough.

@LongJohnSilver All I can see is the embedded video, and I can't really spot the issue. Can you make the original MTS file available for download on your Vimeo page?

@LPowell ok thanks. I use Edius, I got it just editing the file properties :)