asarnow
commented
11 months ago I've determined that this issue occurs if just directories were specified during prediction. Denoised tomograms look normal using explicit lists of even and odd inputs.
Open asarnow opened 11 months ago
asarnow
commented
11 months ago I've determined that this issue occurs if just directories were specified during prediction. Denoised tomograms look normal using explicit lists of even and odd inputs.
On the left is the original tomogram (with default 3dmod smoothing) and on the right the denoised version. I used the default parameters from the front page here for training, and the pixel size 5.48 A. For prediction the tile size is [2,4,2]. I'm confused by the appearance of the transverse membrane-like feature at about a 45 deg. angle in the center of the denoised tomogram. If real, it's essentially invisible in the original, yet more strongly denoised than any of features that are visible. Could it be an artifact of some kind? Also, do you have any advice on selecting parameters to reduce the smoothing effect? Thanks!
PS to clarify many of my denoised tomograms have similar new features which appear similar to but transverse to real features. Here is a second example, where the new feature is apparently a hole edge, complete with a fan shaped reconstruction artifact.
Images on the left are 2 denoised tomo slices, on the right is the original.
PPS The even and odd tomograms were produced using AreTomo to reconstruct even and odd summed tilt series using alignments from the full sums. The even and odd tomograms look about the same as expected, no hint of the "sideways" feature in any of non-denoised image.