LV-MHP-v2

MHP v2.0 数据集包含 25,403 张图像，每张图像至少有两个人（平均为 3 个）。我们随机选择 5,000 张图像及其相应的注释作为测试集。其余的形成一个包含 15,403 个图像的训练集和一个包含 5,000 个图像的验证集。对于每个实例，定义和注释了 58 个语义类别，除了“背景”类别，即“帽子/帽子”、“头盔”、“脸”、“头发”、“左臂”、“右臂”、 “左手”、“右手”、“防护服”、“比基尼/文胸”、“夹克/风衣/连帽衫”、“T 恤”、“马球衫”、“毛衣”、“单品” ”、“躯干皮肤”、“裤子”、“短裤/游泳短裤”、“裙子”、“丝袜”、“袜子”、“左靴”、“右靴”、“左鞋” ”, “右鞋”, “左高跟鞋”, “右高跟鞋”, “左凉鞋”, “右凉鞋”, “左腿”, “右腿”, “左脚”, “右脚”、“外套”、“连衣裙”、“长袍”、“连身裤”、“其他全身衣服”、“头饰”、“背包”、“球”、“球棒”、“腰带” ”、“瓶子”、“手提包”、“箱子”、“太阳镜”、“眼镜”、“手套”、“围巾”、“雨伞”、“钱包/钱包”、“手表”、“腕带”、“领带” ”、“其他配件”、“其他上身衣服”和“其他下身衣服”。每当对应的类别出现在当前图像中时，每个实例都有一组完整的注释。

The MHP v2.0 dataset contains 25,403 images, each of which includes at least two persons (averaging 3 persons per image). We randomly select 5,000 images along with their corresponding annotations as the test set. The remaining data are split into a training set with 15,403 images and a validation set with 5,000 images. For each instance, 58 semantic categories are defined and annotated, excluding the "background" category. The specific categories are as follows: "hat/cap", "helmet", "face", "hair", "left arm", "right arm", "left hand", "right hand", "protective clothing", "bikini/bra", "jacket/coat/hoodie", "T-shirt", "polo shirt", "sweater", "single piece", "torso skin", "trousers", "shorts/swim trunks", "skirt", "stockings", "socks", "left boot", "right boot", "left shoe", "right shoe", "left high-heeled shoe", "right high-heeled shoe", "left sandal", "right sandal", "left leg", "right leg", "left foot", "right foot", "outerwear", "dress", "robe", "jumpsuit", "other full-body garments", "headwear", "backpack", "ball", "bat", "belt", "bottle", "handbag", "box", "sunglasses", "eyeglasses", "gloves", "scarf", "umbrella", "wallet/purse", "watch", "wristband", "tie", "other accessories", "other upper-body garments", and "other lower-body garments". Each instance has a complete set of annotations whenever the corresponding category appears in the current image.

搜集汇总

背景与挑战

背景概述

LV-MHP-v2是一个用于多人人体解析的数据集，包含25,403张图像，每张图像平均有3个人，并精细标注了58个语义类别，涵盖身体部位、服装和配件。数据集划分为15,403张训练图像、5,000张验证图像和5,000张测试图像，适用于计算机视觉任务如姿势估计和人体分析。该数据集由多个研究机构于2018年发布，支持相关学术研究。