liveBook · Manning

Atrous spatial pyramid pooling

As explained in Fight Fraud with Machine Learning

Atrous spatial pyramid pooling (ASPP) is a multiscale context module that applies parallel atrous convolutions at multiple rates to capture features with different effective fields of view.

Atrous spatial pyramid pooling (ASPP) is a multiscale context module used in DeepLabV3 segmentation. It applies parallel atrous convolutions at multiple rates and effective fields of view, capturing features at different scales without reducing image resolution. section 6.2

In document forgery detection, ASPP supports pixel-level analysis by:

  • Capturing manipulated receipt regions that vary in size. section 6.2
  • Supporting segmentation, where each image pixel is classified as forged or genuine. section 6.5
  • Working alongside 1x1 convolutions that consolidate information across learned feature channels. section 6.2

Receipt forgery segmentation

A DeepLabV3 model can segment a receipt into forged and genuine pixels. ASPP helps the model recognize forgery artifacts across different spatial scales, while weighted cross-entropy addresses the scarcity of forged pixels. section 6.2 section 6.5

Figure 6.8 Architecture of the DeepLabV3 segmentation model that uses atrous convolution, 1x1 convolution, and atrous spatial pyramidal pooling as some distinct features that help in accurate image segmentation. (source: https://learnopencv.com/deeplabv3-ultimate-guide/)

Use the DeepLabV3 architecture figure to locate ASPP alongside atrous and 1x1 convolution layers. section 6.2

How Fight Fraud with Machine Learning covers it

How deep it goes
Explains parallel rates, effective fields of view, and resolution preservation, but does not specify ASPP parameter values.
Examples
Uses forged-receipt segmentation and architectural diagrams comparing ASPP with atrous and 1x1 convolutions.
What it assumes
Assumes familiarity with CNNs and image segmentation while positioning ASPP within a practical forgery-detection workflow.
What you can do after
Apply it via DeepLabV3 to model multiscale features in pixel-level document-forgery detection.
How it connects
Connects atrous convolution, DeepLabV3, 1x1 convolution, image segmentation, and class imbalance.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Fight Fraud with Machine Learning. The section links point to the parts of the book it draws on.