ImageNet Inception v3¶
DeepOBS test problem class for the Inception version 3 architecture on ImageNet.
Details about the architecture can be found in the original paper.
There are many changes from the paper to the official Tensorflow implementation as well as the model.txt that can be found in the sources of the original paper. We chose to implement the version from Tensorflow (with possibly some minor changes)
In the original paper they trained the network using:
- Batch size
- RMSProp with a decay of
0.9and \(\epsilon = 1.0\).
- Initial learning rate
- Learning rate decay every two epochs with exponential rate of
- Gradient clipping with threshold 2.0
- batch_size (int) -- Batch size to use.
- weight_decay (float) -- Weight decay factor. Weight decay (L2-regularization)
is used on the weights but not the biases.
The DeepOBS data set class for ImageNet.
A tensorflow operation initializing the test problem for the training phase.
A tensorflow operation initializing the test problem for evaluating on training data.
A tensorflow operation initializing the test problem for evaluating on test data.
A tf.Tensor of shape (batch_size, ) containing the per-example loss values.
A scalar tf.Tensor containing a regularization term.
A scalar tf.Tensor containing the mini-batch mean accuracy.
Set up the Inception v3 test problem on ImageNet.