Video noise is a known problem in cine cameras or cameras with external RAW recording using BMD VideoAssist or ATMOS Ninja etc.
However, each camera sensor has a particular noise pattern to the recorded RAW files.
The ideal scenario will be a future version of Topaz Video AI with a drop-down menu with a list of different cine camera models for the Topaz engine to optimize the de-noise based on that particular camera model.
The best file format to train the Topaz engine will be 16 bit uncompressed TIFF files, after a direct export from an editing software (without adjustments), at native camera resolution.
I believe 2 seconds (50-60 frames in total will be enough).
For the recorded content, ideally, it will require capturing from a very stable tripod, a static frame without motion.
I believe that content will be ideal for training the topaz engine to chroma/luma changes.
For users familiar with BMD Resolve, create a timeline at your camera native resolution (4/6/8K), debayer at native resolution at RAW tab, and then add a node to colorspace transform to REC.709/Gamma 2.4 or to REC.2020 ST.2084.
Then render as TIFF uncompressed 16-bit RGB Fullrange, zip these frames and upload to Topaz Team.
After some time, Topaz team will have content from a different camera, during day/night/mid-day/cloudy) to train/optimize the engine for specific camera profiles.
