如何使用 OptionsDataset 构建卷积神经网络

Question

This summer I started learning about neural networks and I am currently planning to implement a convolutional neural network for object detection using voc dataset from tensorflow_datasets. I have loaded the dataset using the following code snippet:

ds, info = tfds.load('voc', split=['train','test'], with_info=True, shuffle_files=True)

Then I split the dataset into train and test, accordingly:

train, test = ds[0], ds[1]

I have also built my model using keras layers API. The problem is that, I do not actually know how to train this dataset using my model. How do I input this dataset to my model? What is the most efficient and logical way?

The types of the train, test split is as follows:

print(type(train))
print(type(test))

Output:

<class 'tensorflow.python.data.ops.dataset_ops._OptionsDataset'>
<class 'tensorflow.python.data.ops.dataset_ops._OptionsDataset'>

Answer 1

Train dataset contains the data you use to train your model. Test usually contains the data you use to evaluate your model. Please note that the actual voc dataset has also a validation split:

So the usual logic is to use train dataset to train and validation dataset to validate against while training. And later on you evaluate the performance of your model using the test dataset. So the structure would like this:

model.fit(train_dataset, epochs=x, validation_data=validation_dataset)

model.evaluate(test_dataset)