Skip to content
This repository has been archived by the owner on Feb 17, 2022. It is now read-only.

Latest commit

 

History

History
85 lines (61 loc) · 3.47 KB

tools.md

File metadata and controls

85 lines (61 loc) · 3.47 KB
Raw

Content

Create Logger

The code is in lib/utils/utils.py

Data Reader

Image Reader

The code is in lib/datatools/abstract/tools.py. The aim of this reader is to read the image based on the opencv or pillow and use data augmentation. The final output shape is [C,H,W]

Introduction

The reader is a class, so you need to create the instance firstly and then use the read function of the instance. We think the advantages are that the instance can keep some useful parameters of format, data augmentation and so on.

The parameters of build the reader are:

  • read_format: string. The reading format of the picture. Users can choose 'opencv' or 'pillow'. Default: 'opencv'
  • channel_num: int. The number of images' channel. Default: 3
  • channel_name: string. The name of each channel. Default: 'rgb'
  • params: the params of the configuration file. Default: None
  • transforms: the data augmentation functions of imgaug. Default: None
  • normalize: bool. whether use the normalization. Default: False
  • mean: list. the mean value of each channel. Default: None
  • std: list. the std value of each channel. Default: None
  • deterministic: bool. the same function of the keyword in imgaug. Default: False

After create the instance, users can use the read function to read the image, the parameters are:

  • name: the absolute path of the image
  • flag: use the torchvision transforms ----> 'torchvision'; use other opensource transforms, like imgaug -----> 'other'. Default: 'other'
  • array_type: the return type of the image. 'tensor' | 'ndarray'. Default: 'tensor'

Example

read_format = 'opencv'
channel_num = 3
channel_name = 'rgb'
name='01.jpg'
image_reader = ImageReader(read_fromat=read_format, channel_num=channel_num)
image = image_reader.read(name)

Video Reader

The code is in lib/datatools/abstract/tools.py.

Introduction

The reader is based on the Image Reader. Like the image reader, the video reader also is a class. Users should create the instance and use the read function. The final output shape is [C,D,H,W]

The parameters of build the reader are:

  • image_loader: the image reader instance
  • params: the params of the configuration file. Default: None
  • transforms: the data augmentation functions of imgaug. Default: None
  • normalize: bool. whether use the normalization. Default: False
  • mean: list. the mean value of each channel. Default: None
  • std: list. the std value of each channel. Default: None

After create the instance, users can use the read function to read the image, the parameters are:

  • frames_list: list. the absolute path of each image.
  • start: the start number base on the frames_list
  • end: the end number base on the frames_list
  • clip_length: the length of each clip. Default: 2
  • step: the step between each frame. Default: 1
  • array_type: the output format. Default: 'tensor'

Example

read_format = 'opencv'
channel_num = 3
channel_name = 'rgb'
frames_list = ['01.jpg', '02.jpg', '03.jpg', '04.jpg']
image_reader = ImageReader(read_fromat=read_format, channel_num=channel_num)
video_reader = VideoReader(image_loader=image_reader)
clip = video_reader.read(frames_list=frames_list, 0, 3)