Error during 'Preview' of .ebsp file about emsphinx HOT 7 CLOSED

from emsphinx.

Hi Marc,

Thanks for the really quick reply! I believe this set was uncompressed.. but before I say that for sure I'll gather another dataset and see if I get the same error.

I do have another series of patterns which I exported (an incredible pain storage wise) and will use your Matlab code to generate a .data file.

Let me play around with that and gather another uncompressed .ebsp dataset and report back.

Thanks, Marc!

Ben

from emsphinx.

Hey @marcdegraef

Finally got back to this issue. I have rescanned a couple datasets with uncompressed patterns. Same issue is arising with the
"vector subscript out of range" when previewing patterns.

Hopefully this isn't too complicated of a fix..

Thanks for all of your help!

Ben

from emsphinx.

Hi Ben,

I managed to get it working by incorporating this pull request. It looks like the height and width of the patterns is switched around in the histogram equalization code. I think the Ni example data has square patterns, so it is not an issue, but the .ebsp files typically have rectangular patterns where this becomes an issue.

I found another issue with pattern batch sizes when working with .ebsp files before I finally got it working, but will raise it as a separate issue. If I make the batch size equal to a factor of the total number of patterns, then it completes the analysis.

Johan

from emsphinx.

Hey Johan,

Thank you for the update! I'll incorporate your pull request and see how it plays with my data. I will report back with any findings.

Cheers,
Ben

from emsphinx.

Hi Ben,

I managed to get it working by incorporating this pull request. It looks like the height and width of the patterns is switched around in the histogram equalization code. I think the Ni example data has square patterns, so it is not an issue, but the .ebsp files typically have rectangular patterns where this becomes an issue.

I found another issue with pattern batch sizes when working with .ebsp files before I finally got it working, but will raise it as a separate issue. If I make the batch size equal to a factor of the total number of patterns, then it completes the analysis.

Johan

Hey Johan,

Solution worked for me but towards the end of analysis I get another error. Is this what you have seen as well?
Cheers,

from emsphinx.

Hi Ben,

I have created another issue for this error: #13

You have 255x196 = 44100 patterns in the .ebsp file. Under "Indexing parameters" on the right, try setting the "batchsize" to say 100, which is a factor of the total number of patterns. At the moment it is set to 0, which will automatically determine the batchsize. This number might not be a factor of the total number of patterns. Some initial debugging shows that this error might be related to reading the last partial batch from the .ebsp in the following loop of the index_ebsd.cpp file.

	size_t batches = idxData.pat->numPat() / nml.batchSize;//how many batches are needed
	if(batches * nml.batchSize < idxData.pat->numPat()) ++batches;//extra batch for leftovers
	for(size_t i = 0; i < batches; i++) {//loop over batches
		const size_t start = i * nml.batchSize;//first pattern
		const size_t end = std::min(start + nml.batchSize, idxData.pat->numPat());//last pattern
		pool.schedule(std::bind(idxData.workItem, std::placeholders::_1));//queue indexing
	}

Please also make sure to output the vendor file as ".ang" and not ".ctf" as it will give another error. I have logged this as a separate issue #14. You might also find the discussions on the pattern center determination on the Oxford system useful.

from emsphinx.