In this line (below), it seems that the code uses `masks` to predict boxes, https://github.com/microsoft/X-Decoder/blob/165f8a6314ac84f5c36aaab7216f90dd97e38a43/modeling/architectures/xdecoder_model.py#L922 but in [line 913](https://github.com/microsoft/X-Decoder/blob/165f8a6314ac84f5c36aaab7216f90dd97e38a43/modeling/architectures/xdecoder_model.py#L913), the predicted boxes are already obtained. https://github.com/microsoft/X-Decoder/blob/165f8a6314ac84f5c36aaab7216f90dd97e38a43/modeling/architectures/xdecoder_model.py#L913 Why do not use predicted bboxes directly?
In this line (below), it seems that the code uses
masksto predict boxes,X-Decoder/modeling/architectures/xdecoder_model.py
Line 922 in 165f8a6
but in line 913, the predicted boxes are already obtained.
X-Decoder/modeling/architectures/xdecoder_model.py
Line 913 in 165f8a6
Why do not use predicted bboxes directly?