search

ImageMonkey / imagemonkey-core

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

Browse mode feature request - random and unlabelled #250

Open dobkeratops opened 5 years ago

dobkeratops commented 5 years ago

Similar to #203

It would be useful to have a way to search for unlabelled images,

Suggestion 1: some special keywords that work in the search bar: “unlabelled”, “random”,”uploads”(the most recently uploaded images),”history” (the images you have visited recently),”unannotated”,”new labels” (images containing a label that doesn’t yet have any annotations)

Suggestion 2: if you hit go with an empty request, fill it with one of those, and carry out that search

Unified mode and add labels could have a different default?

..in unified mode, initialise it to a truly random set (then you only bother entering a specific search term if you didn’t see any interesting annotate life images straight away)

.. In add labels mode , the default could be unlabelled (ie the images in most urgent need of labelling..)

Could unlabelled +random be a default for both? (Ie show unlabelled images first, then if there aren’t any, show random images?.. or a random mix (some examples of each special case listed above ?)

Eventually these could be put into some dedicated UI? I suppose you could have those in a drop down selector.

Would it be possible to combine a count of the current labels and annotations on the image to produce a priority score ?(priority = -number of labels - % annotated pixels - % annotated labels - (number of existing examples of each label, to prioritise images with new labels)

bbernhard commented 5 years ago

short update: It's now possible to query the database for unlabeled images. See:

Selection_025

The "order by random" will follow :)

dobkeratops commented 5 years ago

That’s awesome ! Works fine . I’d been killing time doing manual scrapes .. theres 20000 unlabelled images now. there’s quite a bit of variety scattered between a lot of samey examples.. so it’ll be great to go back and browse to pick out the images that stand out

bbernhard commented 5 years ago

Yeah, we are already at 70k images - that's really awesome! Many, many thanks for all the contributions and the hard work you put into! It's really great to see the dataset grow!

btw: The free labeling is also almost done (a few minor issues & improvements left) and will go live soon. After that, I'll probably back off a bit from developing new features, as I need to prepare the server move. I'll move the site to a different hosting provider where we get more disk space (we are currently at ~60% disk space usage) and more power. :)