caffe源码学习-io

image detect label file

1,create lmdb dataset
python scripts/create_annoset.py -h
执行结果为：

usage: create_annoset.py [-h] [–redo] [–anno-type ANNO_TYPE]
[–label-type LABEL_TYPE] [–backend BACKEND]
[–check-size] [–encode-type ENCODE_TYPE]
[–encoded] [–gray]
[–label-map-file LABEL_MAP_FILE] [–min-dim MIN_DIM]
[–max-dim MAX_DIM] [–resize-height RESIZE_HEIGHT]
[–resize-width RESIZE_WIDTH] [–shuffle]
[–check-label]
root listfile outdir exampledir
Create AnnotatedDatum database
positional arguments:
root The root directory which contains the images and
annotations.
listfile The file which contains image paths and annotation
info.
outdir The output directory which stores the database file.
exampledir The directory to store the link of the database files.
optional arguments:
-h, –help show this help message and exit
–redo Recreate the database.
–anno-type ANNO_TYPE
The type of annotation {classification, detection}.
–label-type LABEL_TYPE
The type of label file format for detection {xml,
json, txt}.
–backend BACKEND The backend {lmdb, leveldb} for storing the result
–check-size Check that all the datum have the same size.
–encode-type ENCODE_TYPE
What type should we encode the image as
(‘png’,’jpg’,…).
–encoded The encoded image will be save in datum.
–gray Treat images as grayscale ones.
–label-map-file LABEL_MAP_FILE
A file with LabelMap protobuf message.
–min-dim MIN_DIM Minimum dimension images are resized to.
–max-dim MAX_DIM Maximum dimension images are resized to.
–resize-height RESIZE_HEIGHT
Height images are resized to.
–resize-width RESIZE_WIDTH
Width images are resized to.
–shuffle Randomly shuffle the order of images and their labels.
–check-label Check that there is no duplicated name/label.

label_type及为对应图片的标注文件的类型