Function eval_sample

t2i_keypoints_eval.py:180–199

Run the full pipeline for one benchmark sample and attach the scores.

(sample: dict, judge: GeminiJudge)

def eval_sample(sample: dict, judge: GeminiJudge) -> dict:
    """Run the full pipeline for one benchmark sample and attach the scores."""
    prompt = sample["prompt"]
    enhanced_prompt = prompt_enhancer(prompt)
    sample["reprompt"] = enhanced_prompt

    image = text_to_image(enhanced_prompt)

    scores = []
    for key_point, key_point_desc in zip(
        sample["prompt_points"], sample["prompt_points_des"]
    ):
        instruction = build_instruction(
            prompt, key_point, key_point_desc, sample["language"]
        )
        verdict = judge(instruction, image)
        scores.append(parse_verdict(verdict))

    sample["res_score_list"] = scores
    return sample


def report(results: list) -> dict: