Rafael Figueiredo Prudencio

@cem-keskin I think that using tf.where is more susceptible to issues than tf.cond. I just encountered a similar issue where NaNs were being backpropagated when the trace of the rotation...

I'm not sure what are the performance implications of my solution, but I ended up just reshaping the batch dimensions into a single one, applying tf.map_fn to each element in...

Hi! I didn't send any emails, but I'd also like to know more details about the submission. I registered for the competition through the Google forms, but didn't get any...

I forgot to mention, but these results were for an 84x84 depth image, which is the resolution used by the original DQN network.

Any update on this?