what about unprocessed video denoising results

[oneTaken]

I find that you use unprocessed video + PG noise model to generate enough training video pairs, then tuning with the real video dataset.

Considering the train/dev real scenes is 6, and the unprocess method can synthesis much more dataset, what about the results before tuning?

Thx.

[cao-cong]

The scenes between train and test set in MOT Challenge dataset are similar, and the other scenes in train set are not good enough for test. So we didn't synthesize more data and test on it.