File size: 2,523 Bytes
68d7505
da8e0e6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
68d7505
 
da8e0e6
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
{
  "name": "MMHal-Bench",
  "release_date": "2023-12-01",
  "subsets": {
    "main": {
      "language": [
        "en"
      ],
      "modalities": [
        "single_image_start"
      ],
      "task_type": "open_ended_qa",
      "score_pipeline": [
        "llm-judge"
      ],
      "score_params": {
        "rubric": "scale",
        "scale": [
          0,
          6
        ],
        "official_judge_model": "gpt-4",
        "judge_inputs": [
          "image_content"
        ]
      },
      "score_protocol": {
        "reference": "Official eval script https://huggingface.co/datasets/Shengcao1006/MMHal-Bench/blob/main/eval_gpt4.py (lines 94-96: judge template consumes image_content, question, gt_answer, model_answer) — GPT-4 rates each response 0-6 (informativeness x hallucination rubric: 0-2 = hallucination present, 3-6 = none). Not implemented in VLMEvalKit or lmms-eval.",
        "note": "Official headline = average rating (0-6) AND hallucination rate = fraction of responses rated 0-2, with per-question-type breakdown (8 types x 12 questions, 96 rows). Per-sample binary grading deviates. Rows carry the image_content object list (extra) that the official judge template interpolates. Per-row question_type extra carries MMHal categories (attribute/relation/holistic/other...), NOT scorer vocabulary — scorer must not treat it as a question_type signal."
      },
      "prompt_template": "<image>{{ question }}",
      "prompt_template_source": {
        "origin": "official",
        "reference": "https://github.com/open-compass/VLMEvalKit/blob/main/vlmeval/dataset/image_vqa.py (MMHal-Bench — judge-scored; bare question)",
        "notes": "Tier 3: MMHal-Bench judge-scored eval: bare question per the official MMHal-Bench paper."
      },
      "mapping_from_source": {
        "media": {
          "from": "image",
          "type": "list",
          "min_items": 1,
          "max_items": 1
        },
        "id": {
          "from": "image_id"
        },
        "question": {
          "from": "question"
        },
        "answer": {
          "from": "gt_answer",
          "optional": true
        },
        "extra": {
          "question_type": {
            "from": "question_type"
          },
          "image_content": {
            "from": "image_content"
          }
        },
        "source": {
          "format": "json",
          "url": {
            "test": "https://huggingface.co/datasets/Shengcao1006/MMHal-Bench"
          }
        }
      }
    }
  }
}