Some Frequently Asked Questions

  1. 1. Who can participate in the contest?

    The contest is open to everyone. However, opportunities at Microsoft may not be available to participants who are not students.

  2. 2. What programming languages can be used?

    Any programming language of your choice can be used.

  3. 3. For the train data and the test data, will we be given only image features or original images as well?

    You will be provided with both the raw images and feature vectors for each image. You do not necessarily have to work on feature vectors provided by us. You can use any feature extraction algorithm of your choice on the raw images provided to you.

  4. 4. Can you suggest some links for learning Machine learning? It will be very helpful.

    Please visit the Getting Started page on the event website.

  5. 5. Do I need to come to IISc to participate in the contest?

    No, this is an online event; you only need to register here

  6. 6. Can a team have more than two members?

    No, only teams of two members are allowed for the contest.

  7. 7. Can we use training data from an external source in addition to the one provided?

    No, to ensure fairness, this is STRICTLY DISALLOWED.

  8. 8. Can we use codes that are available online? If yes, what if the code we use, might have used additional images to train the model? Won't it violate the clause in answer 7?

    Yes, you can use any resource that is publicly available under an Open Source license (e.g. GPL). In addition to it you can also use MATLAB packages. In case you want to use anything other than this, please notify to us and we will let you know whether you can use it or not. For the second question, the clause is that the input to your algorithm should be the train data provided by us, but it is fine if you use any other code such as edge detector as a subroutine. You do not have to worry about the image data the publicly available code has already used to train its model. Also, you will have to submit your code, along with all the subroutines used by you.

  9. 9. Can we use algorithms already published in papers?

    Yes, you can use algorithms published in papers.

  10. 10. The validation data is very noisy. We are afraid that the validation of our learnt model will not be very good.

    It is true that the validation data is noisy. We have identified and filtered out the noisy images on our side. When you submit your predictions on the validation set we will evaluate it only on the non-noisy images. So don't worry!

  11. 11. How do we extract feature vectors out of the images?

    There are various algorithms for doing this such as SIFT, SURF etc. However, in case you are unfamiliar with this, we will also give feature vectors for each image using a standard algorithm.

  12. 12. Can we use GUI-based tools like Weka, RapidMiner etc? If so, how can we submit the code?

    You can of course use these tools. In that case you will not be able to submit the code obviously, but you should write in your report the sequence of steps you took on the GUI to arrive at your result. As there are some points for the code also, you will lose out on those points. Therefore, even if you are using weka, try to call the libraries of Weka using code rather than using GUI.

  13. 13. Can we use tools used in image vision etc?

    You can use any tool you want. However these should be documented properly when you send us a report on your approach.

  14. 14. Will we get a confirmation after registering for the event?

    Yes, you will get a confirmation mail (to the primary email ID you provided during registration) within 24 hours of registering for the event. If this does not, you might have made a mistake while entering your primary email ID in the registration form - in any case, do revert back to us.

  15. 15. I am unable to log in to the submission portal. What should I do?

    Just mail it to us at

  16. 16. Will we get a participation certificate?

    Yes, provided you submit the final predictions on the test data along with your code and writeup on your approach.

    17. What is the scoring method?

    Your submission will be cross-checked with the true labels. The classification accuracy will be calculated based on the percentage of images your algorithm is able to classify correctly. To decide the winners we will see both the classification accuracy on test data as well as your code.

    18. Can I submit more than once?

    Yes, you can. The latest score will be used.

    19. I did not receive a confirmation email on regsitration. What should I do?

    Try Logging in. If you can login, your registration is successful.

    20. Can I crop the images and then use them?

    You are not allowed to crop the any of the images in the datasets.

    21. While using any GUI tool for object segmentation, is any manual operation on the image allowed? For instance, in GrowCut algorithm, a user vaguely draws some strokes inside the object of interest with an object brush, and outside the object with a background brush. Is this kind of user intervention allowed?

    You cannot use these type of tools. You cannot specify the shape to be learned, by manually drawing it (or seperating foreground from background). The idea is that your algorithm has to automatically learn that.