Use junk masks to improve denoiser training?

Exposure Curation
1

Hi,

For a future version, would it be possible to use the masks from Junk Detector to exclude patches containg junk from training Micrograph Denoiser?

This would be useful in the case of hexafoil grids, where nearly every frame contains gold - I haven’t tested, but I suspect this may not be ideal for training the denoiser to denoise non-junk regions?

Would also be great to use them for excluding junk regions from Patch CTF interpolation - e.g. when collecting data on small hole gold grids with a tilt…

Cheers
Oli

1 Like
2

Or define a central area of the micrograph to train on (perhaps with increase numbers of micrographs to compensate)?

Perhaps a way of weeding out “bad” micrographs might be measure intensity in a, say, 32x32 pixel area in the centre and each corner, and if intensity varies by more than a (user defined?) factor, do not use it for training?

1 Like
3

Hi @olibclarke @rbs_sci thanks! We are planning to make use of the junk masks in the ways you mentioned, essentially for ignoring junk parts of micrographs for all other preprocessing stages.

2 Likes
4

Fantastic, thanks Ali!!

5

Look forward to it reaching public availability!