Hey Ozo, I was participating in your OZOXMAS event and it dawned on me that you're probably doing the verification manually. If we estimate an average of 50 solves per puzzle, that's 3000 image - timestamp verifications (an average of 40 equates to 2400 verifications required).
That sounds dreadfully boring, although if you've already got a solution in mind (e.g. scrolling down a list of submissions with the username, time and screenshot side by side) then it doesn't seem so bad.
With that in mind, I was curious if you wanted an automated solution (or at least, a solution that would probably automate most of the submissions). My idea for the implementation is as follows:
Use HTML/CSS to create "templates" for the username, score and puzzle image. We want to mimic the bold text & background color etc.
Convert these HTML text into an image template using webkit2png to get 3 images for template matching. When implemented correctly, these templates should be exact copies of how the username, score and puzzle image will look like on the page (and by extension, the provided screenshot). Other templates can also be added on to verify the puzzle likeness.
Perform template matching (e.g. with Python OpenCV) with a confidence score; only when all 3 templates are matched in the screenshot with a sufficient confidence value is the automated verification successful. I don't think scale invariance is value for implementation time in this case (since most people would have a browser zoom of 100%), although it is possible.
If the automated verification fails, either the wrong timestamp is input, or the scale of the screenshot is wrong. At this point, manual verification can be performed.
My guess is that this would cut the number of manual verifications you'd have to do by 90% (the baseline being 100% success if everyone puts their times correctly at 100% browser size), although there is some work involved. The usernames of each user (on jigidi) will also need to be stored somewhere for the automated system to use. I can whip up a Python implementation on Google Colab for this template matching thing in my free time if you're interested.
will check it later :)
about easiness i kinda doubt so far :D If submission is valid, I can check it in few seconds. Invalid submissions is where most of time goes. I checked around 700 (first batch, multiple days) submissions manually in under an hour
Thank you Ozo
I will activate it soon. 👍
Edit: it is activated now
Comment has been collapsed.
Hey Ozo, I was participating in your OZOXMAS event and it dawned on me that you're probably doing the verification manually. If we estimate an average of 50 solves per puzzle, that's 3000 image - timestamp verifications (an average of 40 equates to 2400 verifications required).
That sounds dreadfully boring, although if you've already got a solution in mind (e.g. scrolling down a list of submissions with the username, time and screenshot side by side) then it doesn't seem so bad.
With that in mind, I was curious if you wanted an automated solution (or at least, a solution that would probably automate most of the submissions). My idea for the implementation is as follows:
My guess is that this would cut the number of manual verifications you'd have to do by 90% (the baseline being 100% success if everyone puts their times correctly at 100% browser size), although there is some work involved. The usernames of each user (on jigidi) will also need to be stored somewhere for the automated system to use. I can whip up a Python implementation on Google Colab for this template matching thing in my free time if you're interested.
Comment has been collapsed.
sure, your help is appreciated. i know of this functionality but i just don't have time for everything (and my python is not that great :D)
Comment has been collapsed.
I've finished the prototype :)
you can access it below:
https://colab.research.google.com/drive/14GwWqR9wzqv3ZKpA9bEW3KiT1qpriZfv?usp=sharing
HAHA i don't know if you'd find it easier to just manually verify / use it, but the option's there if you want it
feel free to ask me anything
Comment has been collapsed.
will check it later :)
about easiness i kinda doubt so far :D If submission is valid, I can check it in few seconds. Invalid submissions is where most of time goes. I checked around 700 (first batch, multiple days) submissions manually in under an hour
Comment has been collapsed.