I’m curious about the parameter’s meaning.(PSROIPooling)
When I run it for test, it produces an output of size
batch_size * output_dim * pooled_size * pooled_size
If I set output_dim = 8
, this runs well. I’m curious how this function splits an ROI to 8 sections? 2x4 or 4x2 or 1x8? (AFAIK, Position Sensitive ROI should divide an ROI. [1])
And, the parameter named group size
is not clear.
Any documentation or reference will be welcomed.
Thank you for advance.
[1] Li, Yi, et al. “Fully convolutional instance-aware semantic segmentation.” Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition . 2017.