Johnny Lim
Reputation: 5833
Why is the number of features in the 1st layer of CNN for MNIST 32?
Based on the following tutorial:
https://www.tensorflow.org/versions/v0.6.0/tutorials/mnist/pros/index.html#deep-mnist-for-experts
Where is the 32 as the number of features in 1st layer of CNN for MNIST coming from?
And where is the 64 as the number of features in 2nd layer coming from?
Upvotes: 2
Views: 176
Answers (1)
lariven
Reputation: 31
32 and 64 are the number of filters, they are hyper parameters set by hand just like the numbers of neurons of a full connected DNN network.
Upvotes: 0
Related Questions
- How to calculate the number of parameters for convolutional neural network?
- Why are the predictions going wrong with MNIST CNN?
- Wrong predictions with own MNIST-like image data in trained CNN model
- How to load MNIST via TensorFlow (including download)?
- Representing the learned weights of MNIST using Tensorflow graphically
- Oscillating accuracy of CNN training with Tensor Flow for MNIST handwritten digits
- why Test precision is higher than training precision