def inference()

in seamseg/algos/fpn.py [0:0]

• 23 lines of code
• 9 McCabe index (conditional complexity)

    def inference(self, head, x, proposals, valid_size, img_size):
        x = x[self.min_level:self.min_level + self.levels]

        try:
            if proposals.all_none:
                raise Empty

            # Run head on the given proposals
            proposals, proposals_idx = proposals.contiguous
            cls_logits, bbx_logits, _ = self._head(head, x, proposals, proposals_idx, img_size, True, False)

            # Shift the proposals according to the logits
            bbx_reg_weights = x[0].new(self.bbx_reg_weights)
            boxes = shift_boxes(proposals.unsqueeze(1), bbx_logits / bbx_reg_weights)
            scores = torch.softmax(cls_logits, dim=1)

            # Split boxes and scores by image, clip to valid size
            boxes, scores = self._split_and_clip(boxes, scores, proposals_idx, valid_size)

            # Do nms to find final predictions
            bbx_pred, cls_pred, obj_pred = self.bbx_prediction_generator(boxes, scores)

            if bbx_pred.all_none:
                raise Empty

            # Run head again on the finalized boxes to compute instance masks
            proposals, proposals_idx = bbx_pred.contiguous
            _, _, msk_logits = self._head(head, x, proposals, proposals_idx, img_size, False, True)

            # Finalize instance mask computation
            msk_pred = self.msk_prediction_generator(cls_pred, msk_logits)
        except Empty:
            bbx_pred = PackedSequence([None for _ in range(x[0].size(0))])
            cls_pred = PackedSequence([None for _ in range(x[0].size(0))])
            obj_pred = PackedSequence([None for _ in range(x[0].size(0))])
            msk_pred = PackedSequence([None for _ in range(x[0].size(0))])

        return bbx_pred, cls_pred, obj_pred, msk_pred