Multiple output bindings for TRT SCFRD

Hi, I have converted SCFRD_10g.pt to onnx and then to TensorRT. When I run it and check the binding shapes, I more binding outputs than input/output.
```
for binding in fd_engine:
print(fd_engine.get_binding_shape(binding))
if fd_engine.binding_is_input(binding):
print('input')
fd_device_input = cuda.mem_alloc(trt.volume(fd_engine.get_binding_shape(binding)) * fd_engine.max_batch_size * np.dtype(np.float32).itemsize)
else:
print('not input')
fd_host_output = cuda.pagelocked_empty(trt.volume(fd_engine.get_binding_shape(binding)) * fd_engine.max_batch_size, dtype=np.float32)
fd_device_output = cuda.mem_alloc(fd_host_output.nbytes)
```
The output is as follows:
The input size is correct. As for the rest, I expected to have one output, but received many. When I tested conversion (I am using TensorRT 8.0.1) with Yolov5sface, I obtained:
May I check why this is so?

Referring to the link here (https://github.com/SthPhoenix/InsightFace-REST/issues/37), it says that it is anchor center and bbox prediction. Is this correct? And why are there three different values?