Adding way to allow user feedback
Opened this issue · 11 comments
Al slam aalaykum,
Since the detection is sometimes not accurate, why not give the user a way to blur undetected content and unblur falsely blurred content. (And maybe use that input to improve the AI model)
Walaikum Assalm,
Thank you for the suggestion, it's surely a useful feature. How do you suggest the user experience look like? I am thinking maybe right context menu option (right click), do you have other ideas in mind?
Well, I am not really sure what is possible, ideally something like the 3 dots menu on YouTube is very user-friendly, but I am not really sure if that is possible, and it would be too complicated.
On the other hand, the context menu is something I see often utilized on other extensions.
One other way it could be done is element detection, like what adblockers use.
Something else I have seen is that some extensions create a bubble or something like that above and element, which could appear when one hovers over the blurred subject and from there is a drop-down menu or even without a dropdown just toggle on click.
If my description is not clear, I could share screenshots of some examples.
I am a novice dev, if we could break down the task and maybe get some instructions I would be very happy to contribute.
Assalamu alaykum warahmatullahy wabarakatuh
I propose to look at the sponsorblock extension this extension allows users to vote segments where there is a sponsor advert and automatically skips such parts we can do something similar getting the video data and timestamp of the video that was not detected and then get the frames at the timestamp and add that to our dataset and tweak the model we just need the user to click a button that would mark the part of the video by recording timestamps and sending them to an online server
BarakaAllah feekum
It would be great if users could give feedback for the models to develop, I suggest to add a Mouse click while pressing some key for false blur and for moments where it should be blurred and it's not
Thank you all for the input, I surely wanna implement this, however, I am not sure everyone agrees on what's considered Haram. I have had some feedback that says "why does it blur professionally looking women" so some people still mistake it for a nudity only fiter. I (or someone with a similar definition of gaze-lowering) need to go through the users reports and verify them one by one. But, I will consider this if I find a way around this, it's not impossible but surely challenging
We should at least implement a "right click to blur" feature, would be very helpful in my opinion.
Assalamualaikum
I am also a ML Engineer and I am in updating the model for better accuracy.
I genuinely loved this project 🙂😍. Maybe soon i will start work on making mobile app of it...
I must say you re all the gems of Islam.
More powers to everyone...
Waaleikum salaam warahmatullahy wabarakatuh
We should at least implement a "right click to blur" feature, would be very helpful in my opinion.
Yes
it's the perfect solution that prevents people mistakes about haram feedback.
I thought about the idea that some people would give false input based on what they think is haram. What you can do is let the feedback be as choosing between what describes the picture such as
- A woman showing her face and hands only
- A woman showing her hair
- A woman showing above her sleeves...
And then you can build on that what should be blurred and what shouldn't