Caffe:Tour:LayerCatalogue

lastmod 2018-08-17created 2015-11-185 edits

To create a Caffe model you need to define the model architecture in a protocol buffer definition file (prototxt).

Vision Layers

Header: include/caffe/vision_layers.hpp

Vision layers generally take an acquired image as input (INPUT) and produce another image as output (OUTPUT). ¹

Convolution
Pooling
Local Response Normalization (LRN)
im2col

Loss Layers

Loss drives learning by comparing an output to a target and assigning cost to minimize. The loss itself is computed by the forward pass and the gradient w.r.t. to the loss is computed by the backward pass.

Softmax
Sum-of-Squares / Euclidean
Hinge / Margin
Sigmoid Cross-Entropy
Infogain
Accuracy and Top-k

Activation / Neuron Layers

In general, activation / Neuron layers are element-wise operators, taking one bottom blob and producing one top blob of the same size.

ReLU / Rectified-Linear and Leaky-ReLU
Sigmoid
TanH / Hyperbolic Tangent
Absolute Value
Power
BNLL

Data Layers

Data enters Caffe through data layers: they lie at the bottom of nets. Data can come from efficient databases (LevelDB or LMDB), directly from memory, or, when efficiency is not critical, from files on disk in HDF5 or common image formats.

Database
In-Memory
HDF5 Input
HDF5 Output
Images
Windows
Dummy

Common Layers

Inner Product

num_output: the number of filters.
(Probably) the last layer uses this value to classify the labels (Label or Filter). Pay attention to the num_output value in the last layer of train_val.prototxt and deploy.prototxt. ²

layer {
  name: "fc8"
  type: "InnerProduct"
  # ...
  inner_product_param {
    num_output: 1000
    # ...
  }
  # ...
}

Splitting

Flattening

Reshape

Concatenation

Slicing

Elementwise Operations

Argmax

Softmax

Mean-Variance Normalization

Favorite site

References

Original text: Vision layers usually take images as input and produce other images as output. ↩
what is batch_size mean? #585 ↩

sitemap