AttentionWindow

Module that extracts a region from an image based on a top-down position and bottom-up detection data. When the bottom-up data is available, the top-down input will select among these positions. There are a number of possibilities:

Bottom-up only The bottom up location closest to the center will be selected and the size of the attention window will be set by the OBJECT_BOUNDS input if connected. If there is no bottom-up data, the top-down input will be used instead.
Top-down only The top-down location in the image will be selected and the region for the attention window will be set TOP_DOWN_BOUNDS if connected.
Top-down and Bottom-up The top-down position in the image will be used to select the bottom up location closest to it and the size of that object will be used if available.

When no top-down or bottom-up object sizes are available it will be set by the parameter focus_radius or both of focus_width and focus_height.

Example XML definition

A simple example

  <module
      class = "AttentionWindow"
      name = "AttentionWindow"
      output_size = "64"
  />

Parameters

Name	Description	Type	Default
class	The class name for the module; must be "AttentionWindow".	string
name	The name of this instance of the module.	string
window_radius	The radius of the window.	float	15
window_size_x	The width of the window. Calculated from radius if not set.	float	0
window_size_y	The height of the window. Calculated from radius if not set.	float	0
output_size_x	The width of the output.	int	64
output_size_y	The height of the output.	int	64

Module Connections

Inputs

Name	Description
INPUT	The input image
BOTTOM_UP_POSITION	The coordinates of detected objects in the image
BOTTOM_UP_BOUNDS	The bounds of the detected object. Four corners.
BOTTOM_UP_COUNT	The number of detected objects
TOP_DOWN_POSITION	A 2x1 array that represents the center of the focus in the input image. Range from 0 to 1, where 0.5 is the center of the image
TOP_DOWN_BOUNDS	The four corners on the top-down attention window

Outputs

Name	Description
OUTPUT	The size of the output

Author

Christian Balkenius
christian.balkenius@lucs.lu.se
Lund University Cognitive Science

Files

AttentionWindow.h
AttentionWindow.cc
AttentionWindow.ikc