CyberHarem/beidou_genshin

这是一个关于《原神》角色北斗（Beidou）的数据集，包含500张图片及其标签。图片从多个网站（如danbooru、pixiv、zerochan等）爬取，并由DeepGHS团队提供技术支持。数据集的核心标签包括`long_hair, eyepatch, breasts, red_eyes, brown_hair, hair_ornament, hair_over_one_eye, large_breasts, earrings`。此外，README还提供了数据集的下载链接和加载方法，以及标签聚类结果的列表。

原始信息汇总

数据集概述

数据集信息

• 名称: Dataset of beidou/北斗/北斗 (Genshin Impact)
• 描述: 包含500张图片及其标签，涉及角色北斗（Genshin Impact）。
• 核心标签: long_hair, eyepatch, breasts, red_eyes, brown_hair, hair_ornament, hair_over_one_eye, large_breasts, earrings

数据集包

名称	图片数量	大小	类型	描述
raw	500	932.49 MiB	Waifuc-Raw	原始数据，包含元信息（最小边对齐到1400像素，如果更大）。
1200	500	784.65 MiB	IMG+TXT	数据集，短边不超过1200像素。
stage3-p480-1200	1284	1.47 GiB	IMG+TXT	3阶段裁剪数据集，区域不小于480x480像素。

标签聚类结果

原始文本版本

#	样本数	图片示例	标签
0	12		1girl, fur_trim, hair_stick, hairpin, jewelry, solo, upper_body, chinese_clothes, one_eye_covered, smile, cleavage, dress, looking_at_viewer, black_gloves, fingerless_gloves, parted_lips, simple_background, white_background, capelet
1	30		1girl, black_gloves, hairpin, solo, chinese_clothes, fingerless_gloves, hair_stick, fur_trim, one_eye_covered, looking_at_viewer, holding_sword, cleavage, smile, thighhighs, jewelry, red_dress, red_capelet, pelvic_curtain, vision_(genshin_impact), boots, greatsword
2	6		1girl, black_gloves, cleavage, dress, fur_trim, hair_stick, hairpin, looking_at_viewer, one_eye_covered, pelvic_curtain, smile, solo, fingerless_gloves, red_capelet, simple_background, thighs, white_background, chinese_clothes, jewelry, black_thighhighs, boots, vision_(genshin_impact)
3	5		1girl, bare_shoulders, black_bikini, cleavage, looking_at_viewer, navel, solo, collarbone, hairpin, stomach, barefoot, blush, hair_stick, thighs, water, abs, cameltoe, feet, grin, jewelry, muscular_female, ocean, one_eye_covered, toned, wet, white_background
4	17		1girl, blush, hair_stick, hairpin, one_eye_covered, erection, jewelry, navel, futanari, large_penis, looking_at_viewer, nipples, testicles, outdoors, sweat, uncensored, blue_sky, teeth, thighs, veiny_penis, ejaculation, grin, huge_penis, completely_nude, day, huge_breasts, projectile_cum, black_hair, cloud, stomach, 1boy, black_thighhighs, solo_focus

表格版本

#	样本数	图片示例	标签
0	12		X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X
1	30		X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X

搜集汇总

数据集介绍

构建方式

CyberHarem/beidou_genshin数据集的构建方式主要涉及从多个网站（如danbooru、pixiv、zerochan等）抓取北斗（Genshin Impact）角色的图像。这些图像的抓取工作由DeepGHS团队的自动抓取系统完成。数据集包含了500张图像及其对应的标签，图像的核心标签包括长发、眼罩、胸部、红眼、棕色头发、发饰、遮住一只眼睛的头发、大胸、耳环等。这些标签是在数据集中经过筛选和精简的。

使用方法

CyberHarem/beidou_genshin数据集的使用方法相对简单。首先，用户需要从Hugging Face的网站上下载所需的数据包。下载完成后，用户可以使用Waifuc加载原始数据集。Waifuc是一个用于处理图像和标签的工具，它可以帮助用户轻松地处理和查看图像和标签。此外，数据集还提供了多种格式的数据包，包括原始数据包、1200像素数据包和3阶段裁剪数据包，以满足不同用户的需求。用户可以根据自己的需求选择合适的数据包进行使用。

背景与挑战

背景概述

在当今数字艺术创作领域，尤其是游戏角色设计领域，文本到图像的生成技术扮演着越来越重要的角色。CyberHarem/beidou_genshin数据集正是在这样的背景下应运而生，它包含了500张与《原神》中的角色北斗相关的图像及其标签。该数据集的核心标签包括长发、眼罩、胸部、红眼、棕发、发饰、头发遮住一只眼、大胸、耳环等，这些标签均为数据集中所剪裁。这些图像来源于多个网站，如danbooru、pixiv、zerochan等，自动爬取系统由DeepGHS团队提供。数据集的创建旨在推动文本到图像生成技术的发展，特别是针对游戏角色设计的特定需求。

当前挑战

CyberHarem/beidou_genshin数据集的构建面临着一些挑战。首先，在图像收集过程中，需要确保图像的多样性和质量，同时避免版权问题。其次，标签的剪裁和分类需要高度准确，以便于后续的模型训练和使用。此外，由于该数据集的特定性和非普遍性，其应用范围可能受到限制。最后，数据集的安全性也是一个重要的挑战，尤其是在处理涉及敏感内容的情况下。

常用场景

经典使用场景

该数据集经典的使用场景在于文本到图像的生成任务，即根据描述性文本生成对应的图像内容。通过分析数据集中的图像及其标签，研究人员可以训练模型理解文本描述与图像特征之间的关系，从而生成与文本描述高度匹配的图像。此外，由于数据集中包含了一些特定的标签，如`long_hair, eyepatch, breasts`等，这使得该数据集在特定角色的图像生成方面具有独特的价值。

解决学术问题

该数据集解决了文本到图像生成任务中图像与文本描述不匹配的问题。通过提供大量高质量、标签明确的图像，该数据集为研究人员提供了一个良好的训练环境，有助于提高模型生成图像的准确性和真实性。此外，数据集中的标签聚类结果也为研究特定角色或服装风格的图像生成提供了有益的参考。

实际应用

该数据集在实际应用场景中，可以用于游戏开发、虚拟现实、电影制作等领域。例如，游戏开发人员可以利用该数据集生成具有特定角色特征的图像，以丰富游戏内容；虚拟现实开发者可以利用该数据集生成逼真的虚拟场景，提升用户体验；电影制作人员可以利用该数据集生成具有特定风格的图像，以丰富电影视觉效果。

数据集最近研究

最新研究方向

在当前人工智能领域，尤其是文本到图像生成的研究中，CyberHarem/beidou_genshin数据集以其独特的图像内容吸引了研究者的关注。该数据集包含500张与《原神》游戏角色北斗相关的图像，并通过爬取多个网站如danbooru、pixiv、zerochan等获得，为研究者的图像分析和生成任务提供了丰富的素材。在近期的研究中，有研究者利用该数据集进行图像识别和分类，通过分析图像中的标签，如长头发、眼罩、胸围、红眼睛等，来识别和分类图像。此外，还有研究者利用该数据集进行图像生成，通过深度学习模型，如生成对抗网络（GAN），生成与北斗角色相关的图像。这些研究对于提高文本到图像生成技术的准确性和效率具有重要意义，有望推动文本到图像生成技术的发展。