When I click through I only see a 4MB zip file. Where is the dataset?
It's great to have this kind of data available. So much computer vision research is done with a small number of datasets, e.g. imagenet, and there are lots of gaps to fill in on how sell these techniques apply to industrially relevant data like this.
These sort of datasets is not surprising to see. I watch "how it's made" videos on youtube and a lot of factory processes seem to have computer vision based sorting for rejecting rocks, debris, etc.
One of my professors consulted for Del Monte. They had an automated vision system that could spot damaged corn kernels as they flew down the conveyer belt and precisely knock them off into a hopper with a puff of air.
I have a small collection of bean-shaped rocks that I have found over time in my purchases of dry beans. Lesson, rinse and sort your beans! By sort I mean remove a 2-3 shrivelling beans per pound and a rock every 10 lbs.
This is where that other article about one pixel object detector could come into play. Build a detector that maps these good bean properties to the color green and then you have a passive identifier than can check a hundred thousand beans per second.
Put a few parallel lines of these up and you could check a truckload of beans in no time.
Why only 7 type of beans(I haven't looked at the data sets themselves)?.
Frankly there are so many types of dry beans that these should be a informative comprehensive catalogue of information about them available somewhere, but I can't seem to find it.
It's great to have this kind of data available. So much computer vision research is done with a small number of datasets, e.g. imagenet, and there are lots of gaps to fill in on how sell these techniques apply to industrially relevant data like this.