Planet:Amazon from Space in Tensorflow

I’m trying to solve Kaggle’s amazon from space challenge by using Tensorflow. All over the internet everybody is solving the challenge by using keras or some transfer learning model. But I want to solve only by using tensorflow. Can anyone show me any example notebooks of such multi label classification using only tensorflow???

You have to make sure you create correct one hot encoding.
As single image has multiple classes.
So you have to mark all classes to 1.

But what of accuracy and correct prediction??

correct_pred = tf.equal(tf.argmax(logits,1),tf.argmax(y,1))
accuracy = tf.reduce_mean(tf.cast(correct_pred,tf.float32))

Is this still right???