Method forward

models/aios/postprocesses.py:1234–1511

(self, outputs, target_sizes, targets, data_batch_nc, not_to_xyxy=False, test=False)

        
    @torch.no_grad()
    def forward(self, outputs, target_sizes, targets, data_batch_nc, not_to_xyxy=False, test=False):
        # import pdb; pdb.set_trace()
        batch_size = outputs['pred_smpl_beta'].shape[0]
        results = []
        device = outputs['pred_smpl_beta'].device
        for body_model in self.body_model.values():
            body_model.to(device)
        # test with instance num
        # num_select=data_batch_nc['joint_img'][0].shape[0]
        num_select = self.num_select
        out_logits, out_bbox= outputs['pred_logits'], outputs['pred_boxes']

        out_smpl_pose, out_smpl_beta, out_smpl_expr, out_smpl_cam, out_smpl_kp3d, out_smpl_verts = \
            outputs['pred_smpl_fullpose'], outputs['pred_smpl_beta'], outputs['pred_smpl_expr'], \
            outputs['pred_smpl_cam'], outputs['pred_smpl_kp3d'], outputs['pred_smpl_verts']

        out_smpl_kp2d = []
        
        for bs in range(batch_size):
            out_kp3d_i = out_smpl_kp3d[bs]
            out_cam_i = out_smpl_cam[bs]
            out_img_shape = data_batch_nc['img_shape'][bs].flip(-1)[None]
            # out_kp3d_i = out_kp3d_i - out_kp3d_i[:, [0]]
            out_kp2d_i = project_points_new(
                points_3d=out_kp3d_i,
                pred_cam=out_cam_i,
                focal_length=5000,
                camera_center=out_img_shape/2
            )   
            out_smpl_kp2d.append(out_kp2d_i.detach().cpu().numpy())
        out_smpl_kp2d = torch.tensor(out_smpl_kp2d).to(device)

            
        assert len(out_logits) == len(target_sizes)
        assert target_sizes.shape[1] == 2
        
        prob = out_logits.sigmoid()
        topk_values, topk_indexes = \
            torch.topk(prob.view(out_logits.shape[0], -1), num_select, dim=1)
        scores = topk_values
        
        # bbox
        topk_boxes = topk_indexes // out_logits.shape[2]
        labels = topk_indexes % out_logits.shape[2]
        
        if not_to_xyxy:
            boxes = out_bbox
        else:
            boxes = box_ops.box_cxcywh_to_xyxy(out_bbox)

        if test:
            assert not not_to_xyxy
            boxes[:,:,2:] = boxes[:,:,2:] - boxes[:,:,:2]
        
        # gather gt bbox
        boxes_norm = torch.gather(boxes, 1, topk_boxes.unsqueeze(-1).repeat(1,1,4))
        target_sizes = target_sizes.type_as(boxes)
        # from relative [0, 1] to absolute [0, height] coordinates