Very simple idea, maybe not so easy / challenging to implement.
The first person to complete milestone 1 satisfactory, will be awarded the project, because if you can't, then you will definitely not be able to do the rest, therefore I will need some proof before I simply take your word for it and award you the project.
Write a proof of concept application that can recognize a face from at least two approximate angles (front and side view) So when an image is selected, and the algorithm recognizes 'the face', it should indicate the approximate area by drawing frame(s) depending on 1 of 3 levels of accuracy. Not all of the human features (that suggest a face) would always be visible or detectable enough, to accurately identify within an image, therefore the levels should be configurable at run time. You may also add additional settings to tweak the accuracy, at your own discretion.
Level 1 (highest and slowest)
A poly-line along the edges of the features that has been successfully detected.
A circle around each feature.
Level 3 (lowest but fastest)
A rectangle (like the ones some camera's have, or most of the software on the market)
The features that should be attempted to identify includes:
Head (including the hair)
Face (normally distinguished by skin color.
Once you have completed the above assignment, please send me the executable for evaluation (I don't want the source code necessarily, because I won't be paying you necessarily :-)
If you can't do anything besides the rectangle around the head, then you would probably not be able to do the remaining requirements, because there is already software that can do it, and plenty of open source algorithms you may use, and I need your creative abilities / superb algorithmic skills, in order to take facial recognition to the level where it will be useful for me / others. If this is the case, don't send me anything, cos I can copy past too !!
If your attempt is not exactly as expected, but shows that you put in good effort / your ideas show potential, then I will reward you separately or consider adding you to the team.
But only the person with the best results will be officially awarded the project. After payment, I will require the source code, which will be checked for plagiarism, and obviously you may use / build on other peoples source code, but if it's found that you did not do any code of your own, then the payment will be revoked.
Due to the uniqueness of my idea, milestone 2 and 3 will only be discussed with those that prove they have what is needed, and although the budget is small now, the opportunities are endless, all depends on you... and your CODE-FU !!
BONUS *** BONUS *** BONUS
Regardless of the above milestone or project itself, you are all challenged to send me an application (without the source code) to relatively detect female breasts, I will then invite you to my private project, so you can see who the other entrants are.
Detection should be indicated by two squares or preferably circles, and only front view is required for now.
Once invited, you can download the sample files (images with breasts exposed), against which your application / algorithm should be tested. The bidder who's application scores the highest (most number of successful detections) will be awarded the project, but only after you update your bid. The new bid should be a percentage of your previous bid, and that percentage equal to the accuracy of your algorithm tested against the sample set of images. Once updated, I will award you the project.
I will then need to see the source code, in order to determine if you did not cheat by hard coding the areas, and also test against other similar images to determine if your algorithm is not over optimized.
But even if your code is 100% copy pasted, your will still receive the bid amount, but if you actually understand the code, and excited to continue developing, then we can discuss the road going forward