- Description: Performs backpropagation of RoiPoolingWithArgMax. Traverse the pooling result of each ROI and accumulate the backward gradient contributions on the coordinates of the feature map, that is, complete the backward propagation on the entire image.
- Formulas: Wherein: The decision function [i = i^*(r,j)] indicates whether the node i is selected as the maximum value output by the jth output node of the candidate region r.
Each operator has calls. First, [object Object] is called to obtain the input parameters and compute the required workspace size based on the process. Then, [object Object] is called to perform computation.
[object Object]
[object Object]
Parameters
[object Object]Returns:
[object Object]: status code. For details, see .The first-phase API implements input parameter validation. The following error codes may be returned.
[object Object]
- The data types of gradOutput, rois, argmax, and gradInputRef are within the supported range.
- gradOutput, argmax, and gradInputRef have the same data type.
- The shape of gradOutput, argmax, and gradInputRef is 4, and the shape of rois is 2.
- The shape[0] of gradOutput, argmax, and rois is the same.
- The shape[1] of gradOutput and argmax is the same.
- The shape[2] of gradOutput and argmax is equal to pooledH and shape[3] is equal to pooledW.
- The value of rois is greater than or equal to 0.
- pooledH and pooledW are greater than 0.
- rois[:, 1] is less than rois[:, 2] and rois[:, 3] is less than rois[:, 4].
- rois.shape[0] and gradOutput.shape[0] are less than or equal to 1024.
The following example is for reference only. For details, see .
[object Object]