search

ImageMonkey / imagemonkey-core

ImageMonkey is an attempt to create a free, public open source image dataset.
https://imagemonkey.io
47 stars 10 forks source link

how can we label the weight and size? #135

Open bbernhard opened 6 years ago

bbernhard commented 6 years ago

(inspired by this discussion here https://github.com/bbernhard/imagemonkey-trending-labels/issues/79)

I am lately thinking a lot about the calories tracker use case and how we could add the weight and size of food. e.q: 200 grams of lettuice. What's the best way to label that?

Using a single label (i.e 200 grams of lettuice) is probably a bad idea, as we can't use the weight to query for all images where 200 grams of some food are shown. Is it enough to split up the label in two single labels: i.e: 200 grams and lettuice or should we split that up into three labels: 200, grams and lettuice?

If we split it up, how can we make sure that the 200 grams belong to the lettuice and not to some other label?

Can we somehow ensure, that people follow our "guidelines"? Otherwise we might end up with a lot of different labeling styles. I mean we still have some control, as new labels need to be unlocked first in order to become productive (so we could do some minor adjustments), but I guess it would be great if we could "enforce" that somehow?

dobkeratops commented 6 years ago

possible ideas imagine being able to extract <number>grams of <item> as suggested above or <number>grams/<item> (going with the generalised x/y label blending idea..) . it could tell you when it correctly identifies a naming convention,

would there be scope for a generalised numerical quantity system , e.g. with dedicated numerical entry? (imagine being able to attach various adjectives and other qualifiers with dedicated input boxes..

"upload an image with contains:name, mass: ?grams, size: ?metres, properties: ?"

bbernhard commented 6 years ago

would there be scope for a generalised numerical quantity system , e.g. with dedicated numerical entry? (imagine being able to attach various adjectives and other qualifiers with dedicated input boxes..

really good question - also thought about that after I uploaded some photos of the vienna airport. I thought that it would be kinda cool, if city/country labels would be treated differently. That way we could e.q show the country's flag instead of the name in the labels view...I think that would add a nice touch. Similar things could be done with other informations like weather, object dimensions,...

Having such context aware labels in place would be totally awesome and in my opinion also superior to the "everything is a plain label approach", as we could do so much things with it. But I think it's really hard to integrate that without totally killing the flow. With #136 we try to make the labeling process more efficient (hotkey, label delimiters, less switching between keyboard and mouse,...). But in order to support context aware labels we probably would need to add more buttons ("add weather information", "add object dimension").

Another option would maybe be to introduce some sort of pseudo language. e.q:

brick?w=15cm?h=20cm or brick?size=15x20cm?weight=5kg

that way we wouldn't need to switch between keyboard and mouse all the time. with syntax highlighting, autocompletion and syntax checking in place, I think that could work pretty well.