在线体验: https://cheerfun.dev/acg2vec/#Pix2Score
github 主仓库地址( tensorflow 的 savemodel 格式可以在 release 中下载): https://github.com/OysterQAQ/ACG2vec (求 star ~)
基于resnet101对插画的浏览数、收藏数、情色级别的分类预测,以 1e-3 的学习率在动漫插画数据集下进行训练,输入尺寸为 224x224 ,输出字典为
{
"bookmark_predict": {
"0": "0-10",
"1": "10-30",
"2": "30-50",
"3": "50-70",
"4": "70-100",
"5": "100-130",
"6": "130-170",
"7": "170-220",
"8": "220-300",
"9": "300-400",
"10": "400-550",
"11": "550-800",
"12": "800-1300",
"13": "1300-2700",
"14": "2700-∞"
},
"view_predict": {
"0": "0-500",
"1": "500-700",
"2": "700-1000",
"3": "1000-1500",
"4": "1500-2000",
"5": "2000-2500",
"6": "2500-3000",
"7": "3000-4000",
"8": "4000-5000",
"9": "5000-6500",
"10": "6500-8500",
"11": "8500-12000",
"12": "12000-19000",
"13": "19000-35000",
"14": "35000-∞"
},
"sanity_predict": {
"0": "0-2",
"1": "2-4",
"2": "4-6",
"3": "6-7",
"4": "7-∞"
}
}
这是一个专为移动设备优化的页面(即为了让你能够在 Google 搜索结果里秒开这个页面),如果你希望参与 V2EX 社区的讨论,你可以继续到 V2EX 上打开本讨论主题的完整版本。
V2EX 是创意工作者们的社区,是一个分享自己正在做的有趣事物、交流想法,可以遇见新朋友甚至新机会的地方。
V2EX is a community of developers, designers and creative people.