gson error for images having extra dimension of size 1

[karlduderstadt]

First of all, it is really great to have the N5 import and export tools now included in Fiji by default! Thanks very much for this effort! I have been meaning to work with N5 for a while and this makes it much easier.

I have been testing importing and exporting with different images and I keep getting an error from gson when it tries to write an NaN (see below). I seem to get this error from images that have an extra dimension of size 1. For example, an image that contains a Z axis but where all T were collected for Z = 0. Something like X, Y, Z, Channel, Time = 1024, 1024, 1, 2, 160. I think somehow the Z dimension in these videos triggers writing the NaN and gson is set to not allow writing of NaN values, perhaps because they would end up being strings and can cause a lot of problems when parsing json later on...

The videos still seem to be written and I can open them. Should I just ignore this error and continue on? Obviously, another solution would be to always work with the minimal dimensions required - to never have dimensions of size 1. But it seems that the SCIFIO reader we are using opens videos this way with some dimensions of size 1. I don't seem to have the issue when opening the videos with IJ1.

I guess the error below is actually coming from the N5Writer and not really the plugins here, but I thought reporting the issue here would be best since it involves the way videos are opened in Fiji. If you want I can report it over there...

Thanks again for this cool format and helpful plugins. If you don't see this as an issue, I get it, and feel free to close it. But please let me know if you have any tips on how I could deal with this.

[bogovicj]

Thanks for reporting this, and your interest in the plugin / tools!

It looks like you're using the "N5 Viewer" metadata type when exporting - is that correct? One issue is that the N5Viewer metadata format does not support a time axis (officially). If it seems to work, it's an accident :p ( We should document this....)

If the above is correct, could you try the "ImageJ" metadata style? That format does support time and datasets of up to 5 dimensions.

[karlduderstadt]

Thanks for the quick feedback! Yes I had used N5 Viewer mostly, but this error also occurs with the ImageJ metadata style option. Here are the settings I just tested with:

That's interesting about the N5 Viewer mode, I didn't know that. In all cases, it seems to create the datasets. With the settings above, I only get the error near the end. I am still thinking it is related to having a dimension with size 1, but maybe it is something else...

[karlduderstadt]

I haven't figured out what is causing the error, but I think I understand better what you meant by the N5 Viewer doesn't support a time axis.

I was hoping to migrate from HD5 to N5 for use in BDV because of the improved performance in creation and viewing. I guess at the moment, there is no support for N5 in BDV with a time axis. Looking at N5Source I see getSource(...) ignores time and N5Viewer always sets everything to one time point. Is there a plan to support time?

I was thinking about writing my own version of N5Source with a getSource(..) method and uses the time information to give back a resliced RandomAccessibleInterval for one given timepoint using Views.hyperSlice(...). Do you think this would be a reasonable strategy to add time information support for N5 / BDV ? Or would this not be optimal for performance?

At the moment, I was using Spim stuff (SpimSource, etc..) with HD5s. SpimSource has an ImgLoader that takes care of this. I didn't look at how that works yet.