(self, image_tensor)
| 72 | return imgP |
| 73 | |
| 74 | def predict(self, image_tensor): |
| 75 | self.model.eval() |
| 76 | with torch.no_grad(): |
| 77 | image_tensor = image_tensor.to(self.device) |
| 78 | if self.config.Global.loss_type == 'ctc': |
| 79 | outputs = self.model(image_tensor) |
| 80 | else: |
| 81 | pseudo_text = None |
| 82 | outputs = self.model(image_tensor, pseudo_text) |
| 83 | outputs = outputs.softmax(dim=2).detach().cpu().numpy() |
| 84 | preds_str = self.converter.decode(outputs) |
| 85 | |
| 86 | return preds_str |
| 87 | |
| 88 | def __call__(self, image): |
| 89 | image_tensor = self.preprocess(image) |