what is the difference between LISARPNHEAD and STANDARDRPNHEAD

[I3aer]

https://github.com/stevewongv/InstanceShadowDetection/blob/1b75e94394ec0abb8cb39820d6311994efca27cf/projects/LISA/LISA/LISA_rpn.py#L34

`@RPN_HEAD_REGISTRY.register() class LISARPNHead(StandardRPNHead): def init(self, cfg, input_shape: List[ShapeSpec], shadow_object_part=False): super().init(cfg,input_shape)

    if shadow_object_part:
        in_channels = [s.channels for s in input_shape]

        assert len(set(in_channels)) == 1, "Each level must have the same channel!"

        in_channels = in_channels[0]
        self.conv = nn.Conv2d(in_channels , in_channels, kernel_size=3, stride=1, padding=1)

        for l in [self.conv]:
            nn.init.normal_(l.weight, std=0.01)
            nn.init.constant_(l.bias, 0)

def forward(self, features):
    """
    Args:
        features (list[Tensor]): list of feature maps
    """

    pred_objectness_logits = []

    pred_anchor_deltas = []

    for i,x in enumerate(features):

        t = F.relu(self.conv(x))

        pred_objectness_logits.append(self.objectness_logits(t))
        pred_anchor_deltas.append(self.anchor_deltas(t))

    return pred_objectness_logits, pred_anchor_deltas`

I think you duplicate the standard one. Did you check what you implement? The code is terrible. You delete some arguments. Probably, your network is not working. It probably remembers samples. How these papers are accepted to be published :(

[stevewongv]

This code was based on Detectron 0.1. We extended the standard RPN to generate two kinds of region proposals for shadow/object instances and shadow-object associations. Why do you say that we just duplicate the standard one? The standard one is here. This is the first work that proposes to detect shadow/object association. The performance is not very good, but it works on many scenes. I don't see your patient and respect to other people's work.

[I3aer]

You inherit the standard one, class LISARPNHead(StandardRPNHead) This makes them same. Your self.conv is same as that in the standard rpn head. I did not say the originality of your work. As I see, class StandardRPNHead(nn.Module) is same as class LISARPNHead(StandardRPNHead).