Is there a solution for training a model with nested lists in a feature vector?

Question

Python

f_1 = 1
f_2 = 0
X_train = 
y_train = [1, -1]

My data set contains nested lists in f_1 and f_2 variables.

f_1 = [[[1, 2], [2, 3], [3, 4]], [[4, 5], [5, 6], [6, 7]]]

The question is, is it realistic to write, for example, a perceptron for training on just such a data structure?
I tried to aggregate and expand nested features into one vector, the result is bad, although the features are informative (but this is not accurate).
I think there is something similar for working with images from the field of deep learning, but in my case the data is not uniform...