Spaces:
Running
Running
| from pathlib import Path | |
| from typing import Dict | |
| import t2v_metrics | |
| import torch | |
| class VQAMetric: | |
| def __init__(self): | |
| self.device = torch.device( | |
| "cuda" | |
| if torch.cuda.is_available() | |
| else "mps" | |
| if torch.backends.mps.is_available() | |
| else "cpu" | |
| ) | |
| self.metric = t2v_metrics.VQAScore( | |
| model="clip-flant5-xxl", device=str(self.device) | |
| ) | |
| def name(self) -> str: | |
| return "vqa_score" | |
| def compute_score( | |
| self, | |
| image_path: Path, | |
| prompt: str, | |
| ) -> Dict[str, float]: | |
| score = self.metric(images=[str(image_path)], texts=[prompt]) | |
| return {"vqa": score[0][0].item()} | |