why use the torch.asin?

[tonghaiping]

after latitudeDecoder the latitude_field has asin but in the LM , the latitude_field sin return why？

class LatitudeDecoder(nn.Module):
"""Minimal implementation of LatitudeDecoder."""

def __init__(self):
    """Latitude decoder."""
    super().__init__()
    self.decoder = LightHamHead()
    self.linear_pred_latitude = nn.Conv2d(self.decoder.out_channels, 1, kernel_size=1)

def forward(self, data: Dict[str, torch.Tensor]) -> Dict[str, torch.Tensor]:
    """Forward pass."""
    x, log_confidence = self.decoder(data["features"])
    eps = 1e-5  # avoid nan in backward of asin
    lat = torch.tanh(self.linear_pred_latitude(x))
    lat = torch.asin(torch.clamp(lat, -1 + eps, 1 - eps))
    return {"latitude_field": lat, "latitude_confidence": torch.sigmoid(log_confidence)}

def calculate_residuals(
    self, camera: BaseCamera, gravity: Gravity, data: Dict[str, torch.Tensor]
) -> Dict[str, torch.Tensor]:
    """Calculate the residuals for the optimization.

    Args:
        camera (BaseCamera): Optimized camera.
        gravity (Gravity): Optimized gravity.
        data (Dict[str, torch.Tensor]): Input data containing the up and latitude fields.

    Returns:
        Dict[str, torch.Tensor]: Residuals for the optimization.
    """
    perspective_up, perspective_lat = get_perspective_field(camera, gravity)
    perspective_lat = torch.sin(perspective_lat)

    residuals = {}
    if "up_field" in data:
        up_residual = (data["up_field"] - perspective_up).permute(0, 2, 3, 1)
        residuals["up_residual"] = up_residual.reshape(up_residual.shape[0], -1, 2)

    if "latitude_field" in data:
        target_lat = torch.sin(data["latitude_field"])
        lat_residual = (target_lat - perspective_lat).permute(0, 2, 3, 1)
        residuals["latitude_residual"] = lat_residual.reshape(lat_residual.shape[0], -1, 1)

    return residuals