search

clobotics / piou

piou loss and Retail50K dataset
MIT License
191 stars 39 forks source link

The images in train and test csv files are 40063, which is different from 47000 images in paper? #6

Open zhangyuxuan1996 opened 4 years ago

zhangyuxuan1996 commented 4 years ago

Are there val.csv? By the way, the paper pointed out that "we selected half of the original images as the training set, 1/6 as validation set, and 1/3 as the testing set"

zhangyuxuan1996 commented 4 years ago

另外能否解释一下标注？同一张图，第二个字段有些是一个框，有些是多个框，是出于什么样的考虑？ ‘1000050’表示什么？归一化的标注是对应于把原图reshape到(600,800)之后的标注是吗？ 1128186,https://fileman.clobotics.cn/api/file/428711869178469a59ae2d50013a4fc3,"[[0.004991853572295213,0.65368509212730319],[0.87690227753319239,0.6335845896147404],[0.87690227753319239,0.69221105527638194],[0.0027732519846084516,0.708961474036851]]",1,[],0 1128186,https://fileman.clobotics.cn/api/file/428711869178469a59ae2d50013a4fc3,"[[0.88799528547162621,0.017169179229480736],[0.88133948070856594,0.30025125628140703],[0.90130689499774674,0.30025125628140703],[0.90574409817312029,0.37562814070351758],[0.87246507435781884,0.37395309882747069],[0.87246507435781884,0.62688442211055273],[0.88799528547162621,0.62688442211055273],[0.88577668388393938,0.69723618090452266],[0.86359066800707174,0.6989112227805695],[0.87912087912087911,0.99371859296482412],[0.99670676326827745,0.9970686767169179],[0.99448816168059073,0.0020938023450586263]]",**1000050**,[],0 1128188,https://fileman.clobotics.cn/api/file/9181382dd4786737766c66d5ae9a4cdf,"[[0.11892797319932999,0.35050251256281406],[0.983249581239531,0.35217755443886095],[0.98548297040759347,0.38735343383584592],[0.11669458403126745,0.3806532663316583]]",1,[],0

zhangyuxuan1996 commented 4 years ago

It also seems that there are 2987 images in retail50k_train_1.csv that are duplicated in retail50k_train_2.csv?

zhangyuxuan1996 commented 4 years ago

I don’t know if there is any problem with my parsing your .csv data. I feel that there are a lot of missing and mislabeled labels in the annotations. Here are some visual examples http://10.224.29.12:8081/showdirs?dirs=Retail50k