Written Student ID instead of encoding

Added by Arne Vilsen 8 months ago

Hi
I am considering the use of amc for the spring exams, but I was surprised to discover that student IDs have to be encoded by ticks, and not written using numbers.

Am I correct that the tick-encoding is the only way? - is there a reason why it is not implemented using number-recognition? (so I should not try implementing it)

Arne


Replies (5)

RE: Written Student ID instead of encoding - Added by Alexis Bienvenüe 8 months ago

Am I correct that the tick-encoding is the only way?

Yes.

is there a reason why it is not implemented using number-recognition?

Not enough time and skills to code something reliable.

(so I should not try implementing it)

Contributions are very welcome!

RE: Written Student ID instead of encoding - Added by Arne Vilsen 8 months ago

Alexis Bienvenüe wrote:

Am I correct that the tick-encoding is the only way?

Yes.

is there a reason why it is not implemented using number-recognition?

Not enough time and skills to code something reliable.

But names are recognized? - or am I mistaken here?

I thought opencv had standard implementations for such a thing using something like mnist, but I get your point.

(so I should not try implementing it)

Contributions are very welcome!

I opened a 7000 line Perl file... wouldn't know where to start ;)

Perhaps I'll make it before exams in June. For 300 students, I find named sheets unhandy. For a computer-science multiple choice exam, I find student-id tick-encoding unhandy, but it is of course the most reliable.

RE: Written Student ID instead of encoding - Added by Alexis Bienvenüe 8 months ago

But names are recognized?

No. The user has to read the names (the GUI helps her to do this as easily/quickly as possible)

I opened a 7000 line Perl file... wouldn't know where to start ;)

Yes, I'm unfortunately not a programmer.
Can you code something that reads the digits from a box image using OpenCV? You can find some test images here: http://docs.passoire.fr/alexis/amc/local.tgz

RE: Written Student ID instead of encoding - Added by Arne Vilsen 8 months ago

I might be able to look into that. Looking into it, I am not sure OpenCVs classification algorithms are the best solution. One might look into neural networks e.g. tensorflow - I've seen mentions of 97-99% accuracy from sample implementations.

I am not sure if I should be worried about false identification. It might not be a problem since the number read should be found in the list of students. For student-ids containing letters, there might be a problem though, but perhaps there is a data-set available for letters too.

RE: Written Student ID instead of encoding - Added by Arne Vilsen 8 months ago

From a discussion at the coffee machine, it turns out the encoding might not be such a bad idea after all: the have so bad hand-writing that the secretaries can't read the numbers they are writing.

I will create a test-set so they can practice.

Thanks for your quick response.

(1-5/5)