File size: 3,024 Bytes
9d17704
b1db4e3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9d17704
 
b1db4e3
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
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
{
  "name": "EndoBench",
  "release_date": "2026-07-07",
  "subsets": {
    "main": {
      "language": [
        "en"
      ],
      "modalities": [
        "single_image_start"
      ],
      "task_type": "multiple_choice_qa",
      "score_pipeline": [
        "rule-match",
        "llm-match"
      ],
      "score_protocol": {
        "reference": "official@arXiv:2505.23601 / github.com/CUHK-AIM-Group/EndoBench (evaluated via VLMEvalKit) — MCQ evaluation extracts the selected option by string matching first, and if that fails an LLM (GPT) is prompted to extract the chosen option (VLMEvalKit rule-first, LLM-extract fallback).",
        "note": "MCQ over up to 6 options (A-F, variable per row; A/B always present). Gold answer is the option letter. Re-converted 2026-07-07 to restore the options block (source columns A-F) and an MCQ prompt that were missing in the first published copy."
      },
      "prompt_template": "<image>Question: {{ question }}\nOptions:\n{% for k, v in options.items() %}{{ k }}. {{ v }}\n{% endfor %}Please select the correct answer from the options above. \n",
      "prompt_template_source": {
        "origin": "official",
        "reference": "https://github.com/open-compass/VLMEvalKit/blob/main/vlmeval/dataset/image_mcq.py (ImageMCQDataset.build_prompt: Question + 'Options:' block + 'Please select the correct answer from the options above.')",
        "notes": "Tier 3: EndoBench's official repo (github.com/CUHK-AIM-Group/EndoBench) evaluates via VLMEvalKit (run.py --data EndoBench); VLMEvalKit ImageMCQDataset.build_prompt is the canonical MCQ prompt. Dataset authors publish no standalone model-input prompt string."
      },
      "mapping_from_source": {
        "source": {
          "format": "huggingface",
          "url": {
            "test": "https://huggingface.co/datasets/Saint-lsy/EndoBench"
          }
        },
        "id": {
          "from": "index"
        },
        "question": {
          "from": "question"
        },
        "choices": [
          {
            "key": "A",
            "from": "A"
          },
          {
            "key": "B",
            "from": "B"
          },
          {
            "key": "C",
            "from": "C",
            "optional": true
          },
          {
            "key": "D",
            "from": "D",
            "optional": true
          },
          {
            "key": "E",
            "from": "E",
            "optional": true
          },
          {
            "key": "F",
            "from": "F",
            "optional": true
          }
        ],
        "answer": {
          "from": "answer",
          "optional": true,
          "note": "Option letter (A-F)."
        },
        "media": {
          "from": "image",
          "type": "list",
          "min_items": 1,
          "max_items": 1
        },
        "extra": {
          "category": {
            "from": "category"
          },
          "task": {
            "from": "task"
          }
        }
      }
    }
  }
}