DmitryPopov
30th November 2015, 10:05
Hi,
although some of it is old news, I'd like to mention here some free VirtualDub filters developed this year:
1. Film Dirt Cleaner. It's a motion-search-based denoiser that finds similar blocks in both neighbor frames (using motion flow) and if they look more like each other than the current block, it uses them to clamp values inside the current block, eliminating spots and noise. When the frames are too different, an intra-frame degrain method is used to fight little spots. See some before/after animation here:
http://www.infognition.com/dirtcleaner/
The logic is inspired by RemoveDirtMC() from AviSynth that used RemoveDirtSSE2.dll and RemoveGrainSSE2.dll plugins together with mvtools. This VirtualDub filter is a fresh implementation that does not rely on any of those tools or plugins. On multicore PCs it works faster than AviSynth version in my tests, although still not too fast.
Requires VirtualDub 1.9.1 or newer.
2. Intelligent Brightness
This filter automatically adjusts brightness/contrast to bring dynamic range to specified bounds. This is similar to AutoLevels but it does not change white balance (colors remain intact, only brightness changes), and it fights noise amplification and flickering by using different target ranges for dark scenes and using temporal smoothing for its parameters. Target range becomes content-sensitive and you can control it by changing its defining curve in settings dialog.
A little comparison with AutoLevels and MSU Smart Brightness:
https://www.youtube.com/watch?v=zkj5yGLhs8k
Homepage with more descriptions and download:
http://www.infognition.com/brightness/
3. (possibility)
Recently we made a simple one-pass deshaker:
https://www.youtube.com/watch?v=puGY1JJ1Los
http://www.infognition.com/blog/2015/deshaker.html
It buffers up to 50 frames, analyses motion and straightens global motion making it mostly linear. It only deals with translational motion, so it does not fix rotations or in-frame distortions. It's much simpler than Gunnar Thalin's Deshaker however it does its work in single pass which might be important for some folks tired of two-pass processing.
Currently it's part of our SDK but we're thinking of making it a VirtualDub filter. I just need to hear whether it's worth it, do you guys want such a filter?
although some of it is old news, I'd like to mention here some free VirtualDub filters developed this year:
1. Film Dirt Cleaner. It's a motion-search-based denoiser that finds similar blocks in both neighbor frames (using motion flow) and if they look more like each other than the current block, it uses them to clamp values inside the current block, eliminating spots and noise. When the frames are too different, an intra-frame degrain method is used to fight little spots. See some before/after animation here:
http://www.infognition.com/dirtcleaner/
The logic is inspired by RemoveDirtMC() from AviSynth that used RemoveDirtSSE2.dll and RemoveGrainSSE2.dll plugins together with mvtools. This VirtualDub filter is a fresh implementation that does not rely on any of those tools or plugins. On multicore PCs it works faster than AviSynth version in my tests, although still not too fast.
Requires VirtualDub 1.9.1 or newer.
2. Intelligent Brightness
This filter automatically adjusts brightness/contrast to bring dynamic range to specified bounds. This is similar to AutoLevels but it does not change white balance (colors remain intact, only brightness changes), and it fights noise amplification and flickering by using different target ranges for dark scenes and using temporal smoothing for its parameters. Target range becomes content-sensitive and you can control it by changing its defining curve in settings dialog.
A little comparison with AutoLevels and MSU Smart Brightness:
https://www.youtube.com/watch?v=zkj5yGLhs8k
Homepage with more descriptions and download:
http://www.infognition.com/brightness/
3. (possibility)
Recently we made a simple one-pass deshaker:
https://www.youtube.com/watch?v=puGY1JJ1Los
http://www.infognition.com/blog/2015/deshaker.html
It buffers up to 50 frames, analyses motion and straightens global motion making it mostly linear. It only deals with translational motion, so it does not fix rotations or in-frame distortions. It's much simpler than Gunnar Thalin's Deshaker however it does its work in single pass which might be important for some folks tired of two-pass processing.
Currently it's part of our SDK but we're thinking of making it a VirtualDub filter. I just need to hear whether it's worth it, do you guys want such a filter?