How to train a network in Keras to change release size

Question

I have a basic neural network created using Keras. I successfully train a network with data vectors and corresponding output data, which are a vector with two elements. It represents the coordinate (x, y). So the array is in the array.

The problem is that I cannot use training data, where one input vector must correspond to many coordinates. In fact, I want the coordinate vector to be the result, without first knowing the number of coordinates.

Network is being created

model = Sequential()
model.add(Dense(20, input_shape=(196608,)))
model.add(Dense(10))
model.add(Dense(2))

and model summary show output sizes for each layer

_________________________________________________________________
Layer (type)                 Output Shape              Param #
=================================================================
dense_1 (Dense)              (None, 20)                3932180
_________________________________________________________________
dense_2 (Dense)              (None, 10)                210
_________________________________________________________________
dense_3 (Dense)              (None, 2)                 22
=================================================================

, 2. None . , ( )?

Answer 1

An rnn , . , . , 100 . , , . , . - , , ( , , 100).

r-cnn - , . kerasaion, github https://github.com/yhenon/keras-frcnn.

Answer 2

(RNN) , seq2seq, . , LSTM GRU, . , " ", RNN.

keras, , , this one.