What makes this a hard problem for computer vision?
-
by sbi6udr
This task seems to be a textbook case where computer vision should be able to easily solve the problem, especially if you could remove the background pictures (which I assume are decoration/context for humans) or even deliver the lines as pre-parsed lines instead of a bitmap (given the "overshoot" we see on some images, I assume the image is derived from some kind of other data). What makes it harder for computers than for humans?
Posted
-
by MWPALMER
I've been wondering that myself. Shouldn't it be straightforward to code whether curved segments intersect in 3 dimensions, beyond some fixed distance from the center?
Posted
-
by cinquemb
I wonder about this too, but I've started to try and automate it for fun, but because i don't know the colors values they use, its hard to make a robust filter to sample those out, before I want to take a stab at the finding the contours and possible intersections defined by the walkthrough given. With knowing those things this would be way easier with the tools out there…
Or maybe this is some IRB approved study and they want to compare human eyes doing the filtering vs computers 😛
originals:
semi-filtered:
If someone gets farther than me, here's the api endpoint to get random images:
https://api.zooniverse.org/projects/higgs_hunter/subjects?limit=10 via GET
and here's the endpoint to post findings:
https://api.zooniverse.org/projects/higgs_hunter/workflows/545b9c57b0c7823037000002/classifications via POST
and headers for the post:
Host: api.zooniverse.org
User-Agent: {your user agent or bot}
Accept: application/json, text/javascript, /; q=0.01
Accept-Language: en-US,en;q=0.5
Accept-Encoding: gzip, deflate
Content-Type: application/x-www-form-urlencoded; charset=UTF-8
X-Zooniverse-Origin: http://www.higgshunters.org
X-Requested-With: XMLHttpRequest
Referer: https://api.zooniverse.org/proxy
Content-Length: {length of body}
Connection: keep-alive
Pragma: no-cache
Cache-Control: no-cache
with a post body like (that needs to be urlencoded and with no newlines:
classification[subject_ids][]=5460d153b0c782277800bd3f&
classification[annotations][0][key]=pickOne&
classification[annotations][0][value][0][x]=your_x_val&
classification[annotations][0][value][0][y]=your_y_val&
classification[annotations][0][value][0][value]=vertex&
classification[annotations][0][value][0][frame]=2&
classification[annotations][0][value][0][how_many]=3&
classification[annotations][0][value][1][x]=your_x_val&
classification[annotations][0][value][1][y]=your_y_val&
classification[annotations][0][value][1][value]=vertex&
classification[annotations][0][value][1][frame]=2&
classification[annotations][0][value][1][how_many]=5-10&
classification[annotations][1][started_at]=Fri, 29 Nov 2014 22:13:00 GMT&
classification[annotations][1][finished_at]=Fri, 29 Nov 2014 22:13:04 GMT&
classification[annotations][2][user_agent]={ user agent or bot}&
classification[annotations][3][image_index]=2&
classification[annotations][4][image_src]=http://zooniverse-static.s3.amazonaws.com/higgs_hunter/subjects/standard/5460d07fb0c7822778004d31_2.png&
classification[annotations][5][workflow]=untitled_workflow
Posted
-
by MWPALMER
I just re-read the 'how you can help' page. I noticed this text: "in doing so, you will also be teaching our computers how to better spot exotic particle events, speeding up the process of future scientific discoveries! To do this Higgs Hunters shows you a combination of simulated and real data. We need to understand what kind of events can be 'detected' using this site, and so we include computer-generated data as well as real data. You'll be told after each classification if it was a simulation."
Thus, there is some element of truth that it is an experiment on volunteers - but in a more holistic way that advances the science. They are interested in finding out what kinds of 'interesting' events are unrecognizable by their current algorithms, so that they can better update their algorithms.
I think this is why the volunteers always seem to want more instructions, and to want to 'get it right' while the project scientists seem to hold back... Because there might be many things naive observers might notice that a trained scientist might habitually overlook. In my own discipline (Botany) this sometimes happens, and the power of #gestalt is not to be underestimated.
This isn't to say that others cannot write code that would outperform what they already have...
Posted
-
by DZM admin
This would be a good question for the "Ask the Science Team" board... I know they check there carefully!
Posted