RoiAlign

Description

Pooling is performed in each region of interest (RoI).

Input

Three inputs:

x: 4D tensor of type float16 or float.

rois: float16 or float. Has shape (num_rois, 4).

batch_indices: int64, with shape (num_rois,).

Output

One output:

y: tensor of the identical data type as input x. shape=(num_rois, C, output_height, output_width).

Property

mode: string. Defaults to avg.

output_height: int. Defaults to 1.

output_width: int. Defaults to 1.

sampling_ratio: int. Number of sampling points of the interpolation algorithm. Defaults to 0.

spatial_scale: float. Spatial sampling rate relative to the input image. Defaults to 1.0.

coordinate_transformation_mode: string: whether to offset the input value. The default value is half_pixel. This attribute is supported in Opset v16 and later versions.

Constraints

The input of float64 is not supported when the atc tool parameter --precision_mode=must_keep_origin_dtype is used.

ONNX Opset Support

Opset v10/v11/v12/v13/v14/v15/v16/v17/v18