I have finally managed to perfect some NeatVideo presets to get rid of compression artifacts, luma and chroma noise. The results are very encouraging. Detail is preserved beautifully even in high iso shots. Noise banding and posterization from blocking artifacts practically disappears. I recommend using it in conjuntion with my picture style (posted elsewhere on this forum) for best results. You should also add a temporal 3 frame filter to avoid dancing detail. I'm running neatvideo on VirtualDub, but it should work with other versions, such as Premiere, Vegas and After Effects. Ideally, you should work with the original files and not recompressed files (the blocking pattern could change and the filter might get confused). The filter is meant to process the frames directly from their mov originals. You can also add a little noise to the final output so your images look less "plastic". The results can be rendered out to you favorite codec, although I recommend a lossless codec to preserve the render quality.

Now I can shoot intentionally underexposed footage confidently :) Attached, you will find a zip file with preset files (noise profile and denoise settings) for neatvideo.

Cheers.

Thanks Luis.... I'll try them out this weekend..

ive been using it for years, but the problem with the 5d mkII is that noise removal is based on an ISO noise profile, and so you need to lock down your ISO on the camera so its not jumping around, because if you take a noise sample in neatvideo on a ISO 3200 frame then thats the noise profile it works with throughout your clip no matter how many times ISO flips each minute of recording. i had asked the neatvideo developer, if there was some way to read the ISO every so many frames and adjust based on that, but i guess such things arent in the video metadata. so locking down the ISO is important for good noise removal.

Yeah, to come up with the noise profile I posted, I took a series of clips to try and come up with the best solution to the problem, a compromise that works well at any iso without washing away too much detail. The best was an out of focus gradient at iso 1600. The gradient helps blocking and banding show up in the profile, even if the program warns about non uniform luma in the sample area. This in turn ends up cleaning those uniform gradient areas quite well even at lower iso settings. I chose iso 1600 because it showed a little noise, chroma noise and noise banding, but not too much. It helps clean up noisy footage or even underexposed areas even at lower iso very nicely but doesn't get rid of too much detail. A final bonus is that it reduces (not completely, but it helps a lot) chroma aliasing due to the sensor line skipping problem. Apparently, NeatVideo identifies the chroma aliasing as chroma noise :)

I could post some before/after shots, but nothing beats just playing with it using your own footage. I shot some deliberately underexposed footage to ty it out and while the originals looked like bad video, the cleaned up files had a very filmic texture. I should also point out that no matter which picture style you choose in camera, your sharpness setting should be 0, because neatvideo does a much better job at sharpening the footage than the camera does (much less halo effect). Also, as I pointed out earlier, a little noise (like 3-4% after the neatvideo pass helps the image retain a much more organic look.

Hope you guys like it. :)

One last thing,

I wanted to come up with a good one size fits all neatvideo profile because the workflow I've come up with avoids having to batch process all my footage by processing the whole edited sequence instead. I edit with the originals inside newtek SpeedEdit and when I finish I save out an avi project file, which is a sort of frameserver proxy file that you can open inside VirtualDub. This way I don't waste rendering time and/or space, while preserving the original quality. From there, VirtualDub ingests the edited sequence and applies the neatvideo processing only to the edited sequence, saving me hours upon hours of processing. The only problem I've encountered so far is when I do color correction. Color correction alters banding and noise characteristics in the footage, so you risk having some color corrected shots not cleaned up properly by the process. In any case, it sure beats having to preprocess all footage. At 1-1.5min. per processed second of footage on a core 2 duo machine, some projects would take weeks of preprocessing, but the edited sequence can usually be processed overnight.

So anyway, this is what I've come up with working with my 5d so far. If anyone has any other workflows that they'd like to share, I'm all ears. :)