Based on the numerous feedback we received on this problem, we have now updated the problem statement, and the associated dataset to address the said feedback. In this post, we outline the key changes that have been made and the broad reasoning behind the same.
- The new dataset is generated by rotating the Rubik’s cube around an arbitrarily chosen axes. The said axis is consistent throughout the whole dataset.
- In view of the previous change, the
zRotparameters from the training set have been dropped, as the updated problem statement is to just estimate the
xRot(rotation in 0-360 degrees) across an arbitrarily chosen axis.
- The new dataset has 5,000 images in the training set, and 5,000 images in the test set.
NOTE: Please download the whole dataset again before making further submissions to this problem.
Both these changes, also simplify the problem statement a lot more, as the final prediction has to be made in a continuous space of
[0, 360] instead of
[0, 360] x [0, 360] x [0, 360].
The updated problem formulation also deals with the Gimbal Lock issue in the previous training set. The initial idea was to ensure no state close to Gimbal Lock states are represented in the training set and the test set, but clearly, due to an oversight at my end, the error crept up into the final production dataset. Thanks @bobthebuilder for pointing it out!
We hope you guys have fun trying to solve the updated problem statement. And please feel free to reach out to us in case you have any questions or concerns.
Best of Luck,