CNN_TFRecord

An example of Tensorflow pipeline with TFRecord

Implement Tensorflow tutorial in Jetson Nano, tutorial from https://www.tensorflow.org/tutorials/images/cnn

Since cpu and gpu shared memory in Jetson Nano, it's easily lead to out-of-memory when data and model both in memory. Te preserve memory for model, preprocess data with TFRecord, read a small batch of the dataset into memory.

Using tensorflow 2.1.0

CIFAR10 dataset, CNN classification

Write images into TFRecord, load and establish pipeline with tf.data, without loading whole dataset

Training and validating as the Tensorflow tutorial