Feature Extractors

Intellesis Deep Features

• Entire image as input for pre-trained network.
• Note: If you use the CPU for segmentation with Deep Feature sets, the results can be different on different machines because they are hardware (CPU) dependent.
• Take the output from an intermediate layer of that network as feature vector, e.g. output from layer 3 was processed by preceding layers 1 and 2.
• Deep Features 50: Using layer 2 with reduced feature dimension = 50
• Deep Features 64: Using layer 1 with full feature dimension = 64
• Deep Features 70: Using layer 3 with reduced feature dimension = 70
• Deep Features 128: Using layer 2 with full feature dimension = 128
• Deep Features 256: Using layer 3 with full feature dimension = 256