RoiExtractor

Description

Obtains the Region of Interest (ROI) feature matrix from the feature mapping list.

Input

• features: input feature tensor of type float or float16. The shape dimension is 4.
• rois: input ROI tensor of type float or float16.

Output

y: output tensor of type float or float16. The shape is the same as that of the input features.

Attribute

• finest_scale (optional): The data type is int. The default value is 56, indicating the finest scale.
• roi_scale_factor (optional): The data type is float. The default value is 0, indicating the factor related to ROI scaling.
• spatial_scale (optional): The data type is float. It is a spatial scale factor.
• pooled_height (optional): The data type is int. The default value is 7, indicating the height dimension in the pooling operation.
• pooled_width (optional): The data type is int. The default value is 7, indicating the width dimension in the pooling operation.
• sample_num (optional): The data type is int. The default value is 0, indicating the number of samples in the sampling operation.
• pool_mode (optional): The data type is string. It indicates the pooling mode.
• aligned (optional): The data type is bool. The default value is true, indicating whether to perform the alignment operation.

Constraints

None.

ONNX Opset Support

Opset v8/v9/v10/v11/v12/v13/v14/v15/v16