Table Of Contents

• Description
  • Structure
• Parameters
• Additional resources
• License
• Changelog
• Known issues

Description

The PyramidROIAlign plugin performs ROIAlign operation on the output feature maps from FPN(Feature Pyramid Network). It is used in sampleMaskRCNN.

Structure

This plugin supports the NCHW format. It takes mutiple input: roi, feature_maps from FPN.

roi is the ROI candidates from ProposalLayer. Its shape is [N, rois, 4] where N is the batch_size, rois is the number of ROI candidates and 4 is the number of coordinates.

feature_maps are the output of FPN. In sample MaskRCNN, the model we provide contains 4 feature maps from FPN's different stages.

This plugin generate one output tensor of shape [N, rois, C, pooled_size, pooled_size] where C is the channel of mutiple feature maps from FPN and pooled_size is the height(and width) of the feature area after ROIAlign.

Parameters

This plugin has the plugin creator class PyramidROIAlignPluginCreator and the plugin class PyramidROIAlign.

The following parameters were used to create PyramidROIAlign instance:

Additional resources

The following resources provide a deeper understanding of the PyramidROIAlign plugin:

• MaskRCNN
• FPN

License

For terms and conditions for use, reproduction, and distribution, see the TensorRT Software License Agreement documentation.

Changelog

June 2019 This is the first release of this README.md file.

Known issues

There are no known issues in this plugin.