13274243a0
Keep K3 suite selection and report-schema scoring in bash, merge K3/vision dataset_args into dpv4 yamls, and pin EvalScope at 735d920ee911 with local patches. Co-authored-by: Cursor <cursoragent@cursor.com>
157 lines
12 KiB
JSON
157 lines
12 KiB
JSON
{
|
||
"meta": {
|
||
"pretty_name": "MSR-VTT",
|
||
"dataset_id": "AI-ModelScope/msr-vtt",
|
||
"paper_url": "https://www.microsoft.com/en-us/research/publication/msr-vtt-a-large-video-description-dataset-for-bridging-video-and-language/",
|
||
"tags": [
|
||
"MultiModal",
|
||
"Video",
|
||
"ImageCaptioning"
|
||
],
|
||
"metrics": [
|
||
"Bleu_1",
|
||
"Bleu_2",
|
||
"Bleu_3",
|
||
"Bleu_4",
|
||
"METEOR",
|
||
"ROUGE_L",
|
||
"CIDEr"
|
||
],
|
||
"primary_metric": {
|
||
"name": "cider",
|
||
"aggregation": null,
|
||
"dimensions": {}
|
||
},
|
||
"few_shot_num": 0,
|
||
"eval_split": "validation",
|
||
"train_split": "",
|
||
"subset_list": [
|
||
"default"
|
||
],
|
||
"description": "\n## Overview\n\nMSR-VTT is a large-scale open-domain video captioning benchmark for evaluating video-to-text generation.\nThe native adapter groups records by `video_id`, so multiple annotation rows for one video become one sample\nwith multiple reference captions.\n\n## Task Description\n\n- **Task Type**: Video captioning\n- **Input**: Video clip or URL\n- **Output**: One concise natural-language caption\n- **Domains**: Open-domain video understanding and description\n\n## Evaluation Notes\n\n- Default data source: `AI-ModelScope/msr-vtt` on ModelScope, `validation` split\n- Hugging Face `VLM2Vec/MSR-VTT` remains available by setting `dataset_hub=\"huggingface\"` in TaskConfig\n- Primary metric: **CIDEr**\n- Additional metrics: BLEU-1/2/3/4, METEOR, ROUGE-L\n- Set `extra_params.video_dir` to prefer local media files over URL metadata\n",
|
||
"prompt_template": "Describe the video in one concise sentence.",
|
||
"system_prompt": "",
|
||
"few_shot_prompt_template": "",
|
||
"aggregation": "mean",
|
||
"extra_params": {
|
||
"video_dir": {
|
||
"type": "str",
|
||
"description": "Optional local directory containing MSR-VTT video files.",
|
||
"value": ""
|
||
},
|
||
"video_extension": {
|
||
"type": "str",
|
||
"description": "Optional extension override for local videos, for example \"mp4\".",
|
||
"value": ""
|
||
}
|
||
},
|
||
"sandbox_config": {},
|
||
"category": "vlm"
|
||
},
|
||
"statistics": {
|
||
"total_samples": 497,
|
||
"subset_stats": [
|
||
{
|
||
"name": "default",
|
||
"sample_count": 497,
|
||
"prompt_length_mean": 43,
|
||
"prompt_length_min": 43,
|
||
"prompt_length_max": 43,
|
||
"prompt_length_std": null,
|
||
"target_length_mean": 51.41,
|
||
"multimodal": {
|
||
"has_images": false,
|
||
"has_audio": false,
|
||
"has_video": true,
|
||
"video": {
|
||
"count_total": 497,
|
||
"count_per_sample": {
|
||
"min": 1,
|
||
"max": 1,
|
||
"mean": 1
|
||
},
|
||
"duration": null,
|
||
"resolutions": [],
|
||
"formats": [
|
||
"mp4"
|
||
]
|
||
}
|
||
}
|
||
}
|
||
],
|
||
"prompt_length": {
|
||
"mean": 43,
|
||
"min": 43,
|
||
"max": 43,
|
||
"std": null
|
||
},
|
||
"target_length_mean": 51.41,
|
||
"computed_at": "2026-06-09T14:22:00.862722",
|
||
"multimodal": {
|
||
"has_images": false,
|
||
"has_audio": false,
|
||
"has_video": true,
|
||
"video": {
|
||
"count_total": 497,
|
||
"count_per_sample": {
|
||
"min": 1,
|
||
"max": 1,
|
||
"mean": 1
|
||
},
|
||
"duration": null,
|
||
"resolutions": [],
|
||
"formats": [
|
||
"mp4"
|
||
]
|
||
}
|
||
}
|
||
},
|
||
"sample_example": {
|
||
"data": {
|
||
"input": [
|
||
{
|
||
"id": "36044e4b",
|
||
"content": [
|
||
{
|
||
"text": "Describe the video in one concise sentence."
|
||
},
|
||
{
|
||
"video": "https://www.youtube.com/watch?v=A9pM9iOuAzM",
|
||
"format": "mp4",
|
||
"start": 116.03,
|
||
"end": 126.21
|
||
}
|
||
]
|
||
}
|
||
],
|
||
"target": "[\"a family is having coversation\"]",
|
||
"id": 0,
|
||
"group_id": 0,
|
||
"metadata": {
|
||
"references": [
|
||
"a family is having coversation"
|
||
],
|
||
"subset": "default",
|
||
"dataset_id": "AI-ModelScope/msr-vtt",
|
||
"dataset_hub": "modelscope",
|
||
"video": "https://www.youtube.com/watch?v=A9pM9iOuAzM",
|
||
"start": 116.03,
|
||
"end": 126.21,
|
||
"fps": null,
|
||
"video_id": "video6513",
|
||
"category": 14
|
||
}
|
||
},
|
||
"subset": "default",
|
||
"truncated": false
|
||
},
|
||
"readme": {
|
||
"en": "# MSR-VTT\n\n\n## Overview\n\nMSR-VTT is a large-scale open-domain video captioning benchmark for evaluating video-to-text generation.\nThe native adapter groups records by `video_id`, so multiple annotation rows for one video become one sample\nwith multiple reference captions.\n\n## Task Description\n\n- **Task Type**: Video captioning\n- **Input**: Video clip or URL\n- **Output**: One concise natural-language caption\n- **Domains**: Open-domain video understanding and description\n\n## Evaluation Notes\n\n- Default data source: `AI-ModelScope/msr-vtt` on ModelScope, `validation` split\n- Hugging Face `VLM2Vec/MSR-VTT` remains available by setting `dataset_hub=\"huggingface\"` in TaskConfig\n- Primary metric: **CIDEr**\n- Additional metrics: BLEU-1/2/3/4, METEOR, ROUGE-L\n- Set `extra_params.video_dir` to prefer local media files over URL metadata\n\n\n## Properties\n\n| Property | Value |\n|----------|-------|\n| **Benchmark Name** | `msr_vtt` |\n| **Dataset ID** | [AI-ModelScope/msr-vtt](https://modelscope.cn/datasets/AI-ModelScope/msr-vtt/summary) |\n| **Paper** | [Paper](https://www.microsoft.com/en-us/research/publication/msr-vtt-a-large-video-description-dataset-for-bridging-video-and-language/) |\n| **Tags** | `ImageCaptioning`, `MultiModal`, `Video` |\n| **Metrics** | `Bleu_1`, `Bleu_2`, `Bleu_3`, `Bleu_4`, `METEOR`, `ROUGE_L`, `CIDEr` |\n| **Default Shots** | 0-shot |\n| **Evaluation Split** | `validation` |\n\n\n## Data Statistics\n\n| Metric | Value |\n|--------|-------|\n| Total Samples | 497 |\n| Prompt Length (Mean) | 43 chars |\n| Prompt Length (Min/Max) | 43 / 43 chars |\n\n**Video Statistics:**\n\n| Metric | Value |\n|--------|-------|\n| Total Videos | 497 |\n| Videos per Sample | min: 1, max: 1, mean: 1 |\n| Formats | mp4 |\n\n\n## Sample Example\n\n**Subset**: `default`\n\n```json\n{\n \"input\": [\n {\n \"id\": \"36044e4b\",\n \"content\": [\n {\n \"text\": \"Describe the video in one concise sentence.\"\n },\n {\n \"video\": \"https://www.youtube.com/watch?v=A9pM9iOuAzM\",\n \"format\": \"mp4\",\n \"start\": 116.03,\n \"end\": 126.21\n }\n ]\n }\n ],\n \"target\": \"[\\\"a family is having coversation\\\"]\",\n \"id\": 0,\n \"group_id\": 0,\n \"metadata\": {\n \"references\": [\n \"a family is having coversation\"\n ],\n \"subset\": \"default\",\n \"dataset_id\": \"AI-ModelScope/msr-vtt\",\n \"dataset_hub\": \"modelscope\",\n \"video\": \"https://www.youtube.com/watch?v=A9pM9iOuAzM\",\n \"start\": 116.03,\n \"end\": 126.21,\n \"fps\": null,\n \"video_id\": \"video6513\",\n \"category\": 14\n }\n}\n```\n\n## Prompt Template\n\n**Prompt Template:**\n```text\nDescribe the video in one concise sentence.\n```\n\n## Extra Parameters\n\n| Parameter | Type | Default | Description |\n|-----------|------|---------|-------------|\n| `video_dir` | `str` | `` | Optional local directory containing MSR-VTT video files. |\n| `video_extension` | `str` | `` | Optional extension override for local videos, for example \"mp4\". |\n\n## Usage\n\n### Using CLI\n\n```bash\nevalscope eval \\\n --model YOUR_MODEL \\\n --api-url OPENAI_API_COMPAT_URL \\\n --api-key EMPTY_TOKEN \\\n --datasets msr_vtt \\\n --limit 10 # Remove this line for formal evaluation\n```\n\n### Using Python\n\n```python\nfrom evalscope import run_task\nfrom evalscope.config import TaskConfig\n\ntask_cfg = TaskConfig(\n model='YOUR_MODEL',\n api_url='OPENAI_API_COMPAT_URL',\n api_key='EMPTY_TOKEN',\n datasets=['msr_vtt'],\n dataset_args={\n 'msr_vtt': {\n # extra_params: {} # uses default extra parameters\n }\n },\n limit=10, # Remove this line for formal evaluation\n)\n\nrun_task(task_cfg=task_cfg)\n```\n\n\n",
|
||
"zh": "# MSR-VTT\n\n\n## 概述\n\nMSR-VTT 是一个大规模开放域视频描述（video captioning）基准测试，用于评估视频到文本的生成能力。 \n原生适配器按 `video_id` 对记录进行分组，因此同一视频的多条标注行会被合并为一个样本，并包含多个参考描述。\n\n## 任务描述\n\n- **任务类型**：视频描述（Video captioning）\n- **输入**：视频片段或 URL\n- **输出**：一条简洁的自然语言描述\n- **领域**：开放域视频理解与描述\n\n## 评估说明\n\n- 默认数据源：ModelScope 上的 `AI-ModelScope/msr-vtt`，使用 `validation` 划分\n- 通过在 TaskConfig 中设置 `dataset_hub=\"huggingface\"`，仍可使用 Hugging Face 上的 `VLM2Vec/MSR-VTT`\n- 主要指标：**CIDEr**\n- 其他指标：BLEU-1/2/3/4、METEOR、ROUGE-L\n- 设置 `extra_params.video_dir` 可优先使用本地媒体文件而非 URL 元数据\n\n## 属性\n\n| 属性 | 值 |\n|----------|-------|\n| **基准测试名称** | `msr_vtt` |\n| **数据集ID** | [AI-ModelScope/msr-vtt](https://modelscope.cn/datasets/AI-ModelScope/msr-vtt/summary) |\n| **论文** | [Paper](https://www.microsoft.com/en-us/research/publication/msr-vtt-a-large-video-description-dataset-for-bridging-video-and-language/) |\n| **标签** | `ImageCaptioning`, `MultiModal`, `Video` |\n| **指标** | `Bleu_1`, `Bleu_2`, `Bleu_3`, `Bleu_4`, `METEOR`, `ROUGE_L`, `CIDEr` |\n| **默认示例数量** | 0-shot |\n| **评估划分** | `validation` |\n\n\n## 数据统计\n\n| 指标 | 值 |\n|--------|-------|\n| 总样本数 | 497 |\n| 提示词长度（平均） | 43 字符 |\n| 提示词长度（最小/最大） | 43 / 43 字符 |\n\n**视频统计信息：**\n\n| 指标 | 值 |\n|--------|-------|\n| 视频总数 | 497 |\n| 每样本视频数 | 最小: 1, 最大: 1, 平均: 1 |\n| 格式 | mp4 |\n\n\n## 样例示例\n\n**子集**：`default`\n\n```json\n{\n \"input\": [\n {\n \"id\": \"36044e4b\",\n \"content\": [\n {\n \"text\": \"Describe the video in one concise sentence.\"\n },\n {\n \"video\": \"https://www.youtube.com/watch?v=A9pM9iOuAzM\",\n \"format\": \"mp4\",\n \"start\": 116.03,\n \"end\": 126.21\n }\n ]\n }\n ],\n \"target\": \"[\\\"a family is having coversation\\\"]\",\n \"id\": 0,\n \"group_id\": 0,\n \"metadata\": {\n \"references\": [\n \"a family is having coversation\"\n ],\n \"subset\": \"default\",\n \"dataset_id\": \"AI-ModelScope/msr-vtt\",\n \"dataset_hub\": \"modelscope\",\n \"video\": \"https://www.youtube.com/watch?v=A9pM9iOuAzM\",\n \"start\": 116.03,\n \"end\": 126.21,\n \"fps\": null,\n \"video_id\": \"video6513\",\n \"category\": 14\n }\n}\n```\n\n## 提示模板\n\n**提示模板：**\n```text\nDescribe the video in one concise sentence.\n```\n\n## 额外参数\n\n| 参数 | 类型 | 默认值 | 描述 |\n|-----------|------|---------|-------------|\n| `video_dir` | `str` | `` | 可选的本地目录，包含 MSR-VTT 视频文件。 |\n| `video_extension` | `str` | `` | 可选的本地视频扩展名覆盖，例如 \"mp4\"。 |\n\n## 使用方法\n\n### 使用 CLI\n\n```bash\nevalscope eval \\\n --model YOUR_MODEL \\\n --api-url OPENAI_API_COMPAT_URL \\\n --api-key EMPTY_TOKEN \\\n --datasets msr_vtt \\\n --limit 10 # 正式评估时请删除此行\n```\n\n### 使用 Python\n\n```python\nfrom evalscope import run_task\nfrom evalscope.config import TaskConfig\n\ntask_cfg = TaskConfig(\n model='YOUR_MODEL',\n api_url='OPENAI_API_COMPAT_URL',\n api_key='EMPTY_TOKEN',\n datasets=['msr_vtt'],\n dataset_args={\n 'msr_vtt': {\n # extra_params: {} # 使用默认额外参数\n }\n },\n limit=10, # 正式评估时请删除此行\n)\n\nrun_task(task_cfg=task_cfg)\n```",
|
||
"content_hash": "d07c609aa5a9e1d708fe45b734d34c49",
|
||
"needs_translation": false
|
||
},
|
||
"updated_at": "2026-08-11T15:30:53.596218",
|
||
"translation_updated_at": "2026-07-13T17:06:26"
|
||
}
|