cosine similarity and euclidean distance between features of same sample and features of different sample are almost same in tf.keras

Question

I have extracted features from face and left and right iris using MobileNetV2 in keras tensorflow transfer learning. The features are then concatenated. I have tried to find the