Monday, April 12, 2019 (put topic here)

Attendants: Ulf, Axel, Richard, Malin, Luana, Maren, Josefine, Katharina, Sophia, Subir, Thomas, Michael

Summary about visit to Rheine

(Sophia, Thomas and Richard went)

  • Pictures can be stored internally on computer, later moved to harddrive

  • Problem with putting asparagus through the machine twice (for the whole data set), therefore we will only get labeled data if one of us goes there and does the work

  • we easily ge unlabeled data – we can manually label it ourselves, maybe we can do “decision tree”

  • at one day, we already got ~35.000 unlabeled pictures

to machine:

  • in machine 13 values saved

  • maybe possible to move knife in machine?

  • it seems that it is possible to save the label which the machine puts on it

  • if we can use all values by machine, maybe we don’t even have to use deep learning, but classical methods

  • we can change speed of machine (up to a min. speed)

  • currently 14 categories to sort asparagus

To our work

  • we should compare deep learning (convolutional neural nets) with classical machine learning methods (the net can’t be too deep, because the machine has to be fast)

  • idea: self-supervised learning / semi-supervised learning (= some labeled data)

  • important for sorting:hierarchical division in categories(40-50% of first category)

  • Richard found code to watch folders and move files

Organisatiorial

  • Do we want to use Trello, Asana or other program to organize ourselves?

  • We have here at the IKW computers to where we can store all images

Next steps

  • we need to make description for categories, table of classes + visual description

  • we should build an interface which makes it easy for us to label data ourselves

  • maybe we want to call people from machine/ call company to get labeled data & more information?

  • Everybody should have a look on images/ some data, look at some open cv, try out to apply classical cv methods, get involved and a feeling for the data, what open libraries are out there which we can use?

For next meeting (26.4.)

  • Everybody should present one way how to work with the data

  • Everybody takes one paper & looks what they have done & present it

  • Call Geschäftsstelle (Beate Eibisch)/ Raumverwaltung to ask for a room (get transponders to computer room?)

  • Luana will present Asana

  • Thomas will edit everyone to GitHub

  • Malin makes a doodle when to go to Rheine (24.4., 25.4. or 26.4. https://doodle.com/poll/8bsmthcz9ddu48ci )

  • Ulf ask room for computer visison (only next week) – if it dosnt work Katharina will ask

  • Richard writes service

  • Sophia, Luana look at pictures

  • Josefine, Katha, Michael – look at papers

  • others will still chose

  • maybe we additionally meet Thursday 18.4. at 12:00pm to prepare field trip