(self, img, ori_img_size, track_instances=None)
| 571 | |
| 572 | @torch.no_grad() |
| 573 | def inference_single_image(self, img, ori_img_size, track_instances=None): |
| 574 | if not isinstance(img, NestedTensor): |
| 575 | img = nested_tensor_from_tensor_list(img) |
| 576 | # if track_instances is None: |
| 577 | # track_instances = self._generate_empty_tracks() |
| 578 | track_instances = self._generate_empty_tracks() |
| 579 | |
| 580 | res = self._forward_single_image(img, track_instances=track_instances) |
| 581 | |
| 582 | track_instances = res['track_instances'] |
| 583 | track_instances = self.post_process(track_instances, ori_img_size) |
| 584 | ret = {'track_instances': track_instances} |
| 585 | if 'ref_pts' in res: |
| 586 | ref_pts = res['ref_pts'] |
| 587 | img_h, img_w = ori_img_size |
| 588 | scale_fct = torch.Tensor([img_w, img_h]).to(ref_pts) |
| 589 | ref_pts = ref_pts * scale_fct[None] |
| 590 | ret['ref_pts'] = ref_pts |
| 591 | return ret |
| 592 | |
| 593 | def forward(self, data: dict): |
| 594 | if self.training: |
nothing calls this directly
no test coverage detected