Identifying Super Heroes


I have recently started with the DL course of and it’s astounding & fun! Completed the first two lectures and decided to try with a different database.

CrowdAnalytix is hosting a competition for identifying superheroes! <link here>

This is a multi-class problem (12 classes) with 200-700 images for each class.

I had some limited success on the validation dataset. Using the tips provided in the class I started with an accuracy of 54% but gradually increased to 74% (on validation)

If anyone else has recently joined this course and wants to work on this problem, ping me!



I found this competition a couple of days ago, alas it only has 24 hours left. It’s been fun and a learning experience to work with a real world messy dataset vs the kaggle standard. If you don’t have time to compete, it is still worth joining to take a copy of the dataset for practice.

I’ve not quite been able to crack into the top 10 paying positions (89% vs 91%), but I live in hope that the public leaderboard is full of over fitters.

That’s cool! There are a few more hours remaining for the contest to get over…so good luck! But after the competition gets over, it would be great if you could share some inputs on improving the model.


hello, i heard of this competition. do anyone has a downloaded dataset?

What what? Superheroes?

Enjoy! (Till it lasts!)