gen_proposals.txt

        rpn_outputs = [rpn_head('rpn', pi, cfg.FPN.NUM_CHANNEL, len(cfg.RPN.ANCHOR_RATIOS), fp16=self.fp16)
                       for pi in features]
        multilevel_label_logits = [k[0] for k in rpn_outputs]
        multilevel_box_logits = [k[1] for k in rpn_outputs]
        #multilevel_pred_boxes = [anchor.decode_logits(logits)
                                 #for anchor, logits in zip(multilevel_anchors, multilevel_box_logits)]

        #proposal_boxes, proposal_scores = generate_fpn_proposals(
        #    multilevel_pred_boxes, multilevel_label_logits, image_shape2d)


        multilevel_label_logits = [tf.expand_dims(k[0], 0) for k in rpn_outputs]

        multilevel_box_logits = [tf.transpose(k[1], [2, 3, 0, 1]) for k in rpn_outputs]
        multilevel_box_logits = [tf.reshape(k, (-1, tf.shape(k)[2], tf.shape(k)[3])) for k in multilevel_box_logits]
        multilevel_box_logits = [tf.expand_dims(k, 0) for k in multilevel_box_logits]

        image_shape2d = tf.expand_dims(image_shape2d, 0)
        proposal_boxes, proposal_scores = generate_fpn_proposals_batch_tf_op(multilevel_box_logits,
                                                                             multilevel_label_logits,
                                                                             image_shape2d)
        proposal_boxes = proposal_boxes[:, 1:]
        multilevel_label_logits = [tf.squeeze(k, 0) for k in multilevel_label_logits]
        multilevel_box_logits = [k[1] for k in rpn_outputs]


        if self.training:
            losses = multilevel_rpn_losses(
                multilevel_anchors, multilevel_label_logits, multilevel_box_logits)
        else:
            losses = []

        return BoxProposals(proposal_boxes), losses