put proposals from encoded enc_output.

        Args:
            enc_output (Tensor[batch_size, sequence_length, hidden_size]): Output of the encoder.
            padding_mask (Tensor[batch_size, sequence_length]): Padding mask for `enc_output`.
            spatial_shapes (Tensor[num_feature_levels, 2]): Spatial shapes of the feature maps.

        Returns:
            `tuple(torch.FloatTensor)`: A tuple of feature map and bbox prediction.
                - object_query (Tensor[batch_size, sequence_length, hidden_size]): Object query features. Later used to
                  directly predict a bounding box. (without the need of a decoder)
                - output_proposals (Tensor[batch_size, sequence_length, 4]): Normalized proposals, after an inverse
                  sigmoid.
        r