[Is it a bug?] cfg.BN.USE_PRECISE_STATS True when cfg.DETECTION.ENABLE True

Hey guys, thanks first for the really great work. My problem is about the error when using cfg file with BN.USE_PRECISE_STATS True and cfg.DETECTION.ENABLE True.

According to my understanding, when setting BN.USE_PRECISE_STATS True, the model will conduct the function `calculate_and_update_precise_bn(loader, model, num_iters=200, use_gpu=True)` to update bn's parameters. But inside the function, it uses `model(inputs)` to forward inputs ([code is here](https://github.com/facebookresearch/fvcore/blob/65ac931177327ae3fd11af5d183d3a3743ce4b58/fvcore/nn/precise_bn.py#L74)). But when we using cfg.DETECTION.ENABLE True, the model takes two variable as inputs, it should be like `model(inputs, meta["boxes"])` ([code is here](https://github.com/facebookresearch/SlowFast/blob/95a96634bc7df08e7f8bff1ff362eb903cdb0210/tools/train_net.py#L70)). So assert error occurs inside the RoI align layer:
```

class ROIAlign(nn.Module):
def __init__(self, output_size, spatial_scale, sampling_ratio, aligned=True):
...

def forward(self, input, rois): # rois will be None in function calculate_and_update_precise_bn(...)
assert rois.dim() == 2 and rois.size(1) == 5
return roi_align(
input, rois, self.output_size, self.spatial_scale, self.sampling_ratio, self.aligned
)

```

If I don't have a clear description, please let me know about it. Thx a lot