COCO2014数据集说明
|--coco.name The coco.names file contains a list of 80 object categories in the COCO dataset
|--coco
|--annotations
|--train.json
|--val.json
|--images
|--train2014# 82783 images
|--val2014# 40504 images
|--labels
|--train2014: 82081 text files
|--val2014: 40137 text files
# contain the bounding box coordinates of the objects in the images.
# COCO_val2014_000000000133.txt text file in the val2014 folder
# 59 0.510930 0.442073 0.978141 0.872188 == ID [xc, yc, w, h]
|--trainvalno5k.txt# 117264 images for traing model, and it's also a combination of the images in
# train2014 and val2014 subfolders, except for 5000 images
|--5k.txt# 5000 images that will be used for validation
# 82783 + 40504 = 123287; 122218 = 82081 + 40137; 117264+5000=122264
COCO2014数据集包含80个对象类别,提供训练和验证两部分。训练集有82783张图像,验证集有40504张图像。每个图像的标注文件记录了对象的边界框坐标。此外,trainvalno5k.txt用于训练模型,包含117264张图像,5k.txt则包含5000张用于验证的图像。
585

被折叠的 条评论
为什么被折叠?



