Grid atlas classification training

[Marcus Fislage]

Hi Jonathan,

I am not sure if we are still the only ones testing your software on a JEOL CryoARM, but at least Dirk was now able to fix the setup and we can currently run it without crashing of software or database.

So I tested a bit the procedure and the first obvious problem to me currently is that the evaluation of square quality based on the grid atlas is not great. 

I tested different exposures to see if that can help, but either I get all squares identified as good (even if they have thick ice blobs or cracks) or some setups allowed resulted in good an bad square classification but it was obvious as a human, that this cannot be correct.

I was therefore doubting if that may be related to the training data set that you had compared to the setup of our microscope. I.e. we have the CryoARM and a K3.

Can that be the case and if so, is there a chance to retrain the software over here or with data from here?

Cheers

Marcus

[Jonathan Bouvette]

Hi Marcus,

As of now, you are still the only JEOL user that has SmartScope installed. However, I've met JOEL employees and users at 3DEM GRC and i would expect this to change soon. For the atlas, if you don't min sharing a few screenshots and some of the atlases mrc stacks, we can try to add them to the training set and improve the model. 

How big are the atlases that you're making? If they are very large, like the whole grid, you may want to try smaller ones and see how it goes.

Best,

Jonathan

[Marcus Fislage]

Hi Jonathan,

I made sure to first test your suggestion and made on a new sample a 4x4 and 2x2 grid atlas instead of the full 7x5 grid atlas. The automatic square classification is still underwhelming, so montage size does not seem to matter.

I send you these files separately with a mail containing a download link.

Ignore anything else than the atlas images in there. Hope this will help.

If you have any other idea on what parameter change may need to be tested, let me know.

Cheers

Marcus

--
You received this message because you are subscribed to a topic in the Google Groups "SmartScope" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/smartscope/cjr7DjWNDGk/unsubscribe.
To unsubscribe from this group and all its topics, send an email to smartscope+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/smartscope/c77fa617-0e60-4a98-be7b-960e66cbbdden%40googlegroups.com.

[Jonathan Bouvette]

Hi Marcus,

Thank you for taking the time to acquire these images and sharing the data. I'm downloading the files right now and will run some testing and try to identify a solution.

I'll keep you posted,

Jonathan