text-2-video-human-preferences-runway-alpha

<style>

.vertical-container {
display: flex;
flex-direction: column;
gap: 60px;
}

.image-container img {
height: 150px; /* Set the desired height */
margin:0;
object-fit: contain; /* Ensures the aspect ratio is maintained */
width: auto; /* Adjust width automatically based on height */
}

.image-container {
display: flex; /* Aligns images side by side */
justify-content: space-around; /* Space them evenly */
align-items: center; /* Align them vertically */
}

.container {
width: 90%;
margin: 0 auto;
}

.text-center {
text-align: center;
}

.score-amount {
margin: 0;
margin-top: 10px;
}

.score-percentage {
font-size: 12px;
font-weight: semi-bold;
}

</style>

# Rapidata Video Generation Runway Alpha Human Preference

<a href="https://www.rapidata.ai">
<img src="https://cdn-uploads.huggingface.co/production/uploads/66f5624c42b853e73e0738eb/jfxR79bOztqaC6_yNNnGU.jpeg" width="300" alt="Dataset visualization">
</a>

<a href="https://huggingface.co/datasets/Rapidata/text-2-image-Rich-Human-Feedback">
</a>

<p>
If you get value from this dataset and would like to see more in the future, please consider liking it.
</p>

This dataset was collected in ~1 hour total using the [Rapidata Python API](https://docs.rapidata.ai), accessible to anyone and ideal for large scale data annotation.

# Overview

In this dataset, ~30'000 human annotations were collected to evaluate Runway's Alpha video generation model on our benchmark. The up to date benchmark can be viewed on our [website](https://www.rapidata.ai/leaderboard/video-models).
The benchmark data is accessible on [huggingface](https://huggingface.co/datasets/Rapidata/text-2-video-human-preferences) directly.

# Explanation of the colums

The dataset contains paired video comparisons. Each entry includes 'video1' and 'video2' fields, which contain links to downscaled GIFs for easy viewing. The full-resolution videos can be found [here](https://huggingface.co/datasets/Rapidata/text-2-video-human-preferences/tree/main/Videos).

The weighted_results column contains scores ranging from 0 to 1, representing aggregated user responses. Individual user responses can be found in the detailedResults column.

# Alignment

The alignment score quantifies how well an video matches its prompt. Users were asked: "Which video fits the description better?".

## Examples

<div class="vertical-container">
<div class="container">
<div class="text-center">
<q>A slow-motion scene of artisans carving intricate patterns into a massive wooden sculpture, wood shavings swirling as each precise cut reveals detailed artwork. The steady rhythm of their tools highlights the beauty of the emerging masterpiece.</q>
</div>
<div class="image-container">
<div>
<h3 class="score-amount">Alpha </h3>
<div class="score-percentage">(Score: 90.26%)</div>
<img src="https://assets.rapidata.ai/0088_alpha_2395128211.gif" width=500>
</div>
<div>
<h3 class="score-amount">Ray 2 </h3>
<div class="score-percentage">(Score: 9.74%)</div>
<img src="https://assets.rapidata.ai/0088_ray2_2.gif" width=500>
</div>
</div>
</div>
<div class="container">
<div class="text-center">
<q>A 2D animation of a brave knight riding a dragon over a mystical mountain range. Vibrant colors highlight shimmering clouds and ancient temples below, with the knight’s armor gleaming in the sunlight.</q>
</div>
<div class="image-container">
<div>
<h3 class="score-amount">Alpha </h3>
<div class="score-percentage">(Score: 7.45%)</div>
<img src="https://assets.rapidata.ai/0009_alpha_4110143644.gif" width=500>
</div>
<div>
<h3 class="score-amount">Hunyuan </h3>
<div class="score-percentage">(Score: 92.55%)</div>
<img src="https://assets.rapidata.ai/0009_hunyuan_1724.gif" width=500>
</div>
</div>
</div>
</div>

# Coherence

The coherence score measures whether the generated video is logically consistent and free from artifacts or visual glitches. Without seeing the original prompt, users were asked: "Which video is logically more coherent? E.g. the video where physics are less violated and the composition makes more sense."

## Examples

<div class="vertical-container">
<div class="container">
<div class="image-container">
<div>
<h3>Alpha </h3>
<div class="score-percentage">(Score: 94.27%)</div>
<img src="https://assets.rapidata.ai/0012_alpha_1126960663.gif" width="500" alt="Dataset visualization">
</div>
<div>
<h3>Pika </h3>
<div class="score-percentage">(Score: 5.73%)</div>
<img src="https://assets.rapidata.ai/0012_pika_732944320.gif" width="500" alt="Dataset visualization">
</div>
</div>
</div>
<div class="container">
<div class="image-container">
<div>
<h3>Alpha </h3>
<div class="score-percentage">(Score: 14.21%)</div>
<img src="https://assets.rapidata.ai/0050_alpha_3848581244.gif" width="500" alt="Dataset visualization">
</div>
<div>
<h3>Sora </h3>
<div class="score-percentage">(Score: 85.79%)</div>
<img src="https://assets.rapidata.ai/0050_sora_0.gif" width="500" alt="Dataset visualization">
</div>
</div>
</div>
</div>

# Preference

The preference score reflects how visually appealing participants found each video, independent of the prompt. Users were asked: "Which video do you prefer aesthetically?"

## Examples

<div class="vertical-container">
<div class="container">
<div class="image-container">
<div>
<h3>Alpha </h3>
<div class="score-percentage">(Score: 92.17%)</div>
<img src="https://assets.rapidata.ai/0082_alpha_3251151314.gif" width="500" alt="Dataset visualization">
</div>
<div>
<h3>Hunyuan </h3>
<div class="score-percentage">(Score: 7.82%)</div>
<img src="https://assets.rapidata.ai/0082_hunyuan_1724.gif" width="500" alt="Dataset visualization">
</div>
</div>
</div>
<div class="container">
<div class="image-container">
<div>
<h3>Alpha </h3>
<div class="score-percentage">(Score: 12.73%)</div>
<img src="https://assets.rapidata.ai/0004_alpha_293654896.gif" width="500" alt="Dataset visualization">
</div>
<div>
<h3>Pika </h3>
<div class="score-percentage">(Score: 87.27%)</div>
<img src="https://assets.rapidata.ai/0004_pika_2126399364.gif" width="500" alt="Dataset visualization">
</div>
</div>
</div>
</div>

</br>

# About Rapidata

Rapidata's technology makes collecting human feedback at scale faster and more accessible than ever before. Visit [rapidata.ai](https://www.rapidata.ai/) to learn more about how we're revolutionizing human feedback collection for AI development.

# Other Datasets

We run a benchmark of the major image generation models, the results can be found on our [website](https://www.rapidata.ai/leaderboard/image-models). We rank the models according to their coherence/plausiblity, their aligment with the given prompt and style prefernce. The underlying 2M+ annotations can be found here:

- Link to the [Rich Video Annotation dataset](https://huggingface.co/datasets/Rapidata/text-2-video-Rich-Human-Feedback)
- Link to the [Coherence dataset](https://huggingface.co/datasets/Rapidata/Flux_SD3_MJ_Dalle_Human_Coherence_Dataset)
- Link to the [Text-2-Image Alignment dataset](https://huggingface.co/datasets/Rapidata/Flux_SD3_MJ_Dalle_Human_Alignment_Dataset)
- Link to the [Preference dataset](https://huggingface.co/datasets/Rapidata/700k_Human_Preference_Dataset_FLUX_SD3_MJ_DALLE3)

We have also colleted a [rich human feedback dataset](https://huggingface.co/datasets/Rapidata/text-2-image-Rich-Human-Feedback), where we annotated an alignment score of each word in a prompt, scored coherence, overall aligment and style preferences and finally annotated heatmaps of areas of interest for those images with low scores.

# Rapidata 视频生成 Runway Alpha 人类偏好数据集

<a href="https://www.rapidata.ai">
<img src="https://cdn-uploads.huggingface.co/production/uploads/66f5624c42b853e73e0738eb/jfxR79bOztqaC6_yNNnGU.jpeg" width="300" alt="数据集可视化">
</a>

<p>如果您从本数据集获益并希望未来看到更多同类资源，请考虑为其点赞。</p>

本数据集总计耗时约1小时完成采集，依托[Rapidata Python API](https://docs.rapidata.ai)构建，面向所有用户开放，非常适合大规模数据标注工作。

# 数据集概览

本数据集共收集了约3万条人类标注数据，用于在我们的基准测试中评估Runway的Alpha视频生成模型。最新版基准测试可通过我们的[官网](https://www.rapidata.ai/leaderboard/video-models)查看，基准测试数据可直接在[Hugging Face](https://huggingface.co/datasets/Rapidata/text-2-video-human-preferences)获取。

# 数据集列项说明

本数据集包含成对视频对比样本。每条数据均包含`video1`与`video2`字段，其中存储了用于快速预览的压缩GIF链接。全分辨率视频可通过[此链接](https://huggingface.co/datasets/Rapidata/text-2-video-human-preferences/tree/main/Videos)获取。

`weighted_results`列包含0至1区间的得分，代表汇总后的用户反馈结果。单条用户标注数据可在`detailedResults`列中查看。

# 对齐度

对齐得分用于量化视频与提示词的匹配程度。调研中向用户提出的问题为："哪一段视频更贴合给定的描述？"

## 示例

<div class="vertical-container">
<div class="container">
<div class="text-center">
<q>工匠在巨型木雕上雕刻精细纹路的慢镜头场景：每一次精准下刀都扬起木屑，逐渐显现出细节丰富的艺术品。刀具敲击的稳定节奏凸显出即将完成的杰作之美。</q>
</div>
<div class="image-container">
<div>
<h3 class="score-amount">Alpha </h3>
<div class="score-percentage">(得分：90.26%)</div>
<img src="https://assets.rapidata.ai/0088_alpha_2395128211.gif" width=500>
</div>
<div>
<h3 class="score-amount">Ray 2 </h3>
<div class="score-percentage">(得分：9.74%)</div>
<img src="https://assets.rapidata.ai/0088_ray2_2.gif" width=500>
</div>
</div>
</div>
<div class="container">
<div class="text-center">
<q>一名勇敢骑士骑乘巨龙飞越神秘山脉的2D动画作品：色彩鲜亮的云层与下方的古寺熠熠生辉，骑士的铠甲在阳光下闪耀着光泽。</q>
</div>
<div class="image-container">
<div>
<h3 class="score-amount">Alpha </h3>
<div class="score-percentage">(得分：7.45%)</div>
<img src="https://assets.rapidata.ai/0009_alpha_4110143644.gif" width=500>
</div>
<div>
<h3 class="score-amount">Hunyuan </h3>
<div class="score-percentage">(得分：92.55%)</div>
<img src="https://assets.rapidata.ai/0009_hunyuan_1724.gif" width=500>
</div>
</div>
</div>
</div>

# 连贯性

连贯性得分用于评估生成视频的逻辑自洽性，以及是否存在人工痕迹或视觉瑕疵。在不展示原始提示词的前提下，向用户提出的问题为："哪一段视频的逻辑连贯性更强？例如，物理规则违背更少、画面构图更合理的视频。"

## 示例

<div class="vertical-container">
<div class="container">
<div class="image-container">
<div>
<h3>Alpha </h3>
<div class="score-percentage">(得分：94.27%)</div>
<img src="https://assets.rapidata.ai/0012_alpha_1126960663.gif" width="500" alt="数据集可视化">
</div>
<div>
<h3>Pika </h3>
<div class="score-percentage">(得分：5.73%)</div>
<img src="https://assets.rapidata.ai/0012_pika_732944320.gif" width="500" alt="数据集可视化">
</div>
</div>
</div>
<div class="container">
<div class="image-container">
<div>
<h3>Alpha </h3>
<div class="score-percentage">(得分：14.21%)</div>
<img src="https://assets.rapidata.ai/0050_alpha_3848581244.gif" width="500" alt="数据集可视化">
</div>
<div>
<h3>Sora </h3>
<div class="score-percentage">(得分：85.79%)</div>
<img src="https://assets.rapidata.ai/0050_sora_0.gif" width="500" alt="数据集可视化">
</div>
</div>
</div>
</div>

# 偏好度

偏好度得分用于衡量参与者对视频的视觉吸引力评价，与提示词无关。调研中向用户提出的问题为："从美学角度出发，你更偏好哪一段视频？"

## 示例

<div class="vertical-container">
<div class="container">
<div class="image-container">
<div>
<h3>Alpha </h3>
<div class="score-percentage">(得分：92.17%)</div>
<img src="https://assets.rapidata.ai/0082_alpha_3251151314.gif" width="500" alt="数据集可视化">
</div>
<div>
<h3>Hunyuan </h3>
<div class="score-percentage">(得分：7.82%)</div>
<img src="https://assets.rapidata.ai/0082_hunyuan_1724.gif" width="500" alt="数据集可视化">
</div>
</div>
</div>
<div class="container">
<div class="image-container">
<div>
<h3>Alpha </h3>
<div class="score-percentage">(得分：12.73%)</div>
<img src="https://assets.rapidata.ai/0004_alpha_293654896.gif" width="500" alt="数据集可视化">
</div>
<div>
<h3>Pika </h3>
<div class="score-percentage">(得分：87.27%)</div>
<img src="https://assets.rapidata.ai/0004_pika_2126399364.gif" width="500" alt="数据集可视化">
</div>
</div>
</div>
</div>

<br>

# 关于Rapidata

Rapidata的技术让大规模人类反馈采集工作比以往任何时候都更快捷、更普惠。请访问[rapidata.ai](https://www.rapidata.ai/)了解更多关于我们如何革新AI开发领域的人类反馈采集流程的信息。

# 其他数据集

我们对主流图像生成模型开展了基准测试，测试结果可在我们的[官网](https://www.rapidata.ai/leaderboard/image-models)查看。我们依据模型的连贯性/合理性、与提示词的对齐程度以及风格偏好度对模型进行排名。相关的200万条以上标注数据可通过以下链接获取：

- 链接至[Rich Video Annotation数据集](https://huggingface.co/datasets/Rapidata/text-2-video-Rich-Human-Feedback)
- 链接至[Coherence数据集](https://huggingface.co/datasets/Rapidata/Flux_SD3_MJ_Dalle_Human_Coherence_Dataset)
- 链接至[Text-2-Image Alignment数据集](https://huggingface.co/datasets/Rapidata/Flux_SD3_MJ_Dalle_Human_Alignment_Dataset)
- 链接至[Preference数据集](https://huggingface.co/datasets/Rapidata/700k_Human_Preference_Dataset_FLUX_SD3_MJ_DALLE3)

我们还收集了[丰富人类反馈数据集](https://huggingface.co/datasets/Rapidata/text-2-image-Rich-Human-Feedback)，该数据集对提示词中的每个单词标注了对齐得分，同时对图像的连贯性、整体对齐度与风格偏好进行评分，并最终为低分图像生成了关注区域热力图。