Method forward

detrsmpl/models/architectures/DetrSMPLloss.py:100–228

Args: img (Tensor): Input images of shape (N, C, H, W). Typically these should be mean centered and std scaled. img_metas (list[dict]): A List of image info dict where each dict has: 'img_shape', 'scale_factor', 'flip', and may also co

(self, preds, targets)

        pass

    def forward(self, preds, targets):
        """
        Args:
            img (Tensor): Input images of shape (N, C, H, W).
                Typically these should be mean centered and std scaled.
            img_metas (list[dict]): A List of image info dict where each dict
                has: 'img_shape', 'scale_factor', 'flip', and may also contain
                'filename', 'ori_shape', 'pad_shape', and 'img_norm_cfg'.
                For details on the values of these keys see
                :class:`mmdet.datasets.pipelines.Collect`.
            gt_bboxes (list[Tensor]): Each item are the truth boxes for each
                image in [tl_x, tl_y, br_x, br_y] format.
            gt_labels (list[Tensor]): Class indices corresponding to each box
            gt_bboxes_ignore (None | list[Tensor]): Specify which bounding
                boxes can be ignored when computing the loss.

        Returns:
            dict[str, Tensor]: A dictionary of loss components.
        """
        # super(SingleStageDetector, self).forward_train(img, img_metas)
        # NOTE the batched image size information may be useful, e.g.
        # in DETR, this is needed for the construction of masks, which is
        # then used for the transformer_head.
        pred_pose = preds['pred_pose']
        pred_betas = preds['pred_betas']
        pred_cameras = preds['pred_cameras']
        has_smpl = targets['has_smpl']
        gt_smpl_body_pose = targets[
            'smpl_body_pose']  # [bs_0: [ins_num, 23, 3]]
        gt_smpl_global_orient = targets['smpl_global_orient']
        gt_smpl_body_pose = \
            [torch.cat((gt_smpl_global_orient[i].view(-1, 1, 3),
                        gt_smpl_body_pose[i]), dim=1).float()
             for i in range(len(gt_smpl_body_pose))]
        gt_smpl_betas = targets['smpl_betas']
        gt_smpl_transl = targets['smpl_transl']
        gt_keypoints2d = targets['keypoints2d']
        gt_keypoints3d = targets['keypoints3d']  # [bs_0: [N. K, D], ...]
        img_metas = targets['img_metas']
        if 'has_keypoints3d' in targets:
            has_keypoints3d = targets['has_keypoints3d']
        else:
            has_keypoints3d = None

        if 'has_keypoints2d' in targets:
            has_keypoints2d = targets['has_keypoints2d']
        else:
            has_keypoints2d = None

        img = targets['img']

        batch_input_shape = tuple(img[0].size()[-2:])
        for img_meta in img_metas:
            img_meta['batch_input_shape'] = batch_input_shape

        L, B, N = pred_pose.shape[:3]
        if self.body_model_train is not None:
            pred_output = self.body_model_train(