Network Models
While we were working on a small data, a simple convolutional network was used for classification purpose. This network was trained for some popular of classes and on a particularly targeted dataset. The network was 4 layer deep and was running on epoch=20 . The droupout value was set to 0.5.
Train/Validation accuracy was around 0.85 and results seemed to be quite promising.
Deep convolutional neural networks have led to a series of breakthroughs for image classification. Many other visual recognition tasks have also greatly benefited from very deep models. So, over the years there is a trend to go more deeper, to solve more complex tasks and to also increase /improve the classification/recognition accuracy. But, as we go deeper; the training of neural network becomes 4 difficult and also the accuracy starts saturating and then degrades also. Residual Learning tries to solve both these problems. ResNet50 is a 50 layer deep Residual Network.
For the network:
- inputshape: The input shape in the form (channels, brows, cols)
- num outputs: The number of outputs at final softmax layer
- block fn: The block function to use. This is either ‘basic block‘ or ‘bottleneck‘.
- The original paper used a basic block for layers < 50 repetitions: Number of repetitions of various block units. At each block unit, the number of filters is doubled and the input size is halved
- Both bottleneck and basic residual blocks are supported. To switch them, simply provide the block function here.