Somshubra Majumdar

May I ask why this is necessarily? Se.py contains just one function, which you can copy paste and use in your own models. Why does it need to become a...

Same padding already does this, and adding a zero padding does not increase parameter count. Where was this stated in the paper?

Thats not how you use any RNN. Number of units is not generally the batch size, recurrent activation is not supposed to be changed from sigmoid as it changes the...

I'm not quite sure as to why the performance is significantly lower. I think there is some error in how this one was implemented.

I have my finals now. I won't be able to look at them until the end of the month.

Have you tried a simpler model and seen if it is able to learn on your dataset ? Are the train and validation images from the same datasets ?

Interesting. What type if preprocessing are you using ? Mean std or -1 to 1? Also, are you using regularization default value? Try setting it to 0

Why not try fine tuning one of the pre trained NASNet blocks rather than train from scratch? Since you seem to have significant computation available, I suggest using NASNet Mobile...

Also, use the -1 to 1 preprocessing for NASNets, and especially when using pretrained weights for fine-tuning.

Use NASNet Mobile, not large. Weights for all models are available. Refer to Keras blogpost on Fine-tuning to see how to use no-top models for training.