ABSTRACT Computer vision is the state-of-art of understanding and manipulating images and videos. Disclosure of this work relates to achieve and compare more accurate optimization by measuring the performance analysis of accuracy in vision for classification and for the predictions on TPU using TensorFlow2.0 Keras with GPU. The ability to process large number of features makes Deep Learning models very powerful when dealing with unstructured data. Previous work presented the work extension of testing the optimization growth on vision accuracy of the deep learning models on TPU of the TensorFlow. This work presenting the comparative growth results of using GPU and TPU of TensorFlow. All the results clearly showing great difference between previous works tested on GPU and with TPU the optimized growth performance of vision accuracy of deep learning models using various difficult datasets with the effect of QoS on the TPU of the TensorFlow. Keyword: TPU (Tensor Processing Unit), TensorFlow, Deep learning models, Optimization, Loss, Accuracy.