learning Hierarchical Features for Scene Labeling
Test input: The input into the network was a static image such as the one below:
Training data and desired result: The desired result (which is the same format as the training data given to the network for supervised learning) is an image with large features labelled.
- labeled cows.png
- cow legend.png
Below we can see a flow of the overall approach.
Before being put into the Convolutional Neural Network (CNN) the image is first passed through a Laplacian image processing pyramid to acquire different scale maps. There were three different scale outputs of the image created.
A typical three layer (convolution of kernel with feature map, non-linearity, pooling) CNN architecture was used. The function tanh served as the non-linearity. The kernel being used were 7x7 Toeplitz matrices. The pooling operation was performed by the 2x2 max-pool operator.
The connection weights were applied to all of the images, thus allowing for the detection of scale-invariant features.
For training, stochastic gradient descent was used. To avoid over-fitting, jitter, horizontal flipping, rotations between +8 and -8, and rescaling between 90 and 110% was used.
Unlike previous approaches, the emphasis of this scene-labelling method was to rely on a highly accurate pixel labelling system. So, despite the fact that a variety of approaches were attempted, including SuperPixels, Conditional Random Fields and gPb, the simple approach of super-pixels often yielded state of the art results.