“Just help us classify some logos on https://hunger.openfoodfacts.org/ ” and so people think they help and make things even worse, because of the limits of the robotOFF ai-bot. I learned it the hard way after some trial-and-error and with some limited reading of the developer documentation.
Some German shop brands, as LIDL, are printed on many products in the OFF database, and so they often appear on Hunger Games to be classified by the people. And as the LIDL logo is a brand, it is classified as such.
But if you want to help and later answer some brand-questions, RobotOFF asks you if this product belongs to the brand LIDL and you click yes sometimes, as the logo of the shop is printed on the backside of most LIDL products (see e.g. 20047238), so they belong to this brand.
After some clicking, you may look at the results of the products classified by you and see, that they all had the main brand missing, which is now LIDL, not the real one which is printed on the frontside. And if you read the (old?) documentation on the Wiki, the first brand should be the main brand of the product.
So the LIDL logo should be better classified as shop, but the helpful people have marked over 7000 LIDL logos as brand and there is no tool for fast de- or re-annotation as shop.
Even the manual de-annotation (marking as “no logo”) seems not to work, as I had problems to deselect the last REWE brand logos, which is also a shop, but this time, the logo is part of a complex brand logo which can be much bigger, as for “REWE frei” von (“REWE free of” e.g. 4388860192216) products.
For these big logos, you may sooner or later find out, that they are not recognized by the bot, but maybe there parts which contains the logo of the shop. So people selected the REWE logo, which is part of e.g. “REWE frei von” and marked it as REWE brand instead of “REWE frei von”, which would be better, as the robot would it not recognize it otherwise for now.
So if you find a potential partial logo, try to find out if robotOFF may know the whole one by entering only the product barcode number on https://hunger.openfoodfacts.org/logos/search/ and search for logos the robot knows.
Another problems seem to be, that robotOFF logo recognition takes precedence over text recognized by the generic OCR (Google cloud vision?), so I try to get rid of the remaining REWE shop logos, so that the many “ja!” (“yes!”, e.g. 4388840063376) are recognized, as the logo is to big for the ai-robot and these products have also “Eine Marke von REWE” (“A brand of REWE”) also printed on their front. So the problem is the same as with LIDL.