Friday, May 31, 2019 (Evaluation Hermeler meeting, Feature Extraction)¶

Ulf, Katha, Maren, Sophia, Malin

Talk about consequences of meeting with Hermeler:¶

we can go to Querdell on Thursdays and Wednesdays before lunch, who can go?
we should talk about what we take out of the meeting next time again, when more people are present

we should measure the size of the buckets, where asparagus pieces lie in. Is the picture distorted? Is the middle bucket enough of a reference? Or do we need different measurements for the 3 differnt buckets?
we should measure the size of some asparagus pieces

katha und Richard tried to extract the flower - problematic, not enough information in picture of head
maren: order the labeled data which maren and malin collected the week before - difficult, as time noted down and time in folder does not seem to match –> new approach: try to look at the names
Sophia: width extraction: function is working
Malin: length extraction: function works if asparagus is straight in the picture. For all other cases, a rotation function would be needed in advance. This is probably already done for the curvature? Is it possible to save the coordinates of the uppermost pixel of the asparagus piece?
Michael: curvature extraction? (originally Marens task) Does Michael do rust, too?

how do we want to save/ record our work/ results/ progress? We should note down somewhere what we are doing (continuously), what works or does not work, what differences are, what we comare to the machine
our idea: record intermediate results in a systematic, uniform way. –> a google doc (eg. manuel feature extraction as one chapter of our final report, what works, what does not, what were our approaches, difficulties, successes …. evaluation with comparison which current aproach of machine,…)
we should leave our functions as notebooks, but we should extract he functions into a script, and build a documentation –> sphinx (this is good for program documentation, and you can extract it as pdf/html) OR Tex (advantage here: you can create final documents, and include comments about pictures/graphes)
when creating tasks in asana: we have to keep in mind, that we assign a person to a task, before we solve it, in order to document our work, and so that we can traceback what we did!!! if one does not assign anyone, and solves the task, it seems to be gone forever

download program from computer in Rheine and try to extract features of pictures which we throw in
collect more labeled data! (we need at least 1000 pictures per class, in order to try a neural network approach!) - Malin will go on Monday, but we need more people, for more days (this week and next week and the week after)
drive to Querdell (which is the farm with the newer machine) and collect images there
EVERYONE should document their work until now, their feature extraction funcitons for the final report
responsibiity for organization of documentaition of code - Katha, (will inform herself and tell us guidelines for our documentation in sphinx)
sort the collected labeled images - Maren
create a google doc for our project and start a rough structure with headings and subheadings, begin with documentation of theory which we collected an practical work which we did so far
OR think for yourself of a useful task to do