How to create caffe.deploy from train.prototxt

by

There are two main differences between a “train” prototxt and a “deploy” one:

1. Inputs: While for training data is fixed to a pre-processed training dataset (lmdb/HDF5 etc.), deploying the net require it to process other inputs in a more “random” fashion.
Therefore, the first change is to remove the input layers (layers that push “data” and “labels” during TRAIN and TEST phases). To replace the input layers you need to add the following declaration:

input: "data"
input_shape: { dim:1 dim:3 dim:224 dim:224 }

This declaration does not provide the actual data for the net, but it tells the net what shape to expect, allowing caffe to pre-allocate necessary resources.

2. Loss: the top most layers in a training prototxt define the loss function for the training. This usually involve the ground truth labels. When deploying the net, you no longer have access to these labels. Thus loss layers should be converted to “prediction” outputs. For example, a “SoftmaxWithLoss” layer should be converted to a simple “Softmax” layer that outputs class probability instead of log-likelihood loss. Some other loss layers already have predictions as inputs, thus it is sufficient just to remove them.

Update: see this tutorial for more information.

More Related Contents:

  1. Caffe | solver.prototxt values setting strategy
  2. Why binary_crossentropy and categorical_crossentropy give different performances for the same problem?
  3. Common causes of nans during training of neural networks
  4. How to interpret caffe log with debug_info?
  5. What is `weight_decay` meta parameter in Caffe?
  6. Fine Tuning of GoogLeNet Model
  7. Caffe: What can I do if only a small batch fits into memory?
  8. How to reduce a fully-connected (`”InnerProduct”`) layer using truncated SVD
  9. What is `lr_policy` in Caffe?
  10. What is a `”Python”` layer in caffe?
  11. Many to one and many to many LSTM examples in Keras
  12. Intuitive understanding of 1D, 2D, and 3D convolutions in convolutional neural networks [closed]
  13. Epoch vs Iteration when training neural networks [closed]
  14. How to calculate optimal batch size
  15. What is the difference between loss function and metric in Keras? [duplicate]
  16. Keras Text Preprocessing – Saving Tokenizer object to file for scoring
  17. Calculate the output size in convolution layer [closed]
  18. Does Caffe need data to be shuffled?
  19. Tackling Class Imbalance: scaling contribution to loss and sgd
  20. Is deep learning bad at fitting simple non linear functions outside training scope (extrapolating)?
  21. multi-layer perceptron (MLP) architecture: criteria for choosing number of hidden layers and size of the hidden layer? [closed]
  22. How to interpret loss and accuracy for a machine learning model [closed]
  23. scale the loss value according to “badness” in caffe
  24. What is the difference between Keras model.evaluate() and model.predict()?
  25. Machine Learning Two class classification [closed]
  26. Artificial Neural Network Toplogy
  27. Clarification on a Neural Net that plays Snake
  28. ResNet: 100% accuracy during training, but 33% prediction accuracy with the same data
  29. LSTM module for Caffe
  30. TimeDistributed(Dense) vs Dense in Keras – Same number of parameters

Leave a Comment