Skip to content

Latest commit

 

History

History
executable file
·
35 lines (25 loc) · 1.73 KB

README.md

File metadata and controls

executable file
·
35 lines (25 loc) · 1.73 KB

DataProcessing

The project is for processing dataset, including Cityscapes and PASCAL VOC2012

Usage

Prerequisites:

  1. python 3
  2. numpy
  3. PIL

Visualization using multi-process

It consists of multi-process_visual & pallete

  • run multi-process_visual.py for converting gray predictions to colors.
  • it will use all the cpu are avaliable.
  • pallete.py provides palletes of different datasets, you can custom it yourself.

Converting index of Cityscapes labels

It consists of reverse_idx & cityscapes_labels

  • reverse_idx.py provides two functions for converting the idx.
  • cityscapes_labelsis based on cityscapesScripts

Extra

  • contouris for computing the boundary maps used in pix2pixHD based on instance labels.
  • scriptsis for coping desired images from files and generating lists of dataset (ie. w/ lst, w/o lst)

Update

  • coco2voc.pyconverts coco2017 labels, which are bigger than 1k pixels, to pascal voc format. This scripts based requires pycocotools and pytorch

  • convert_pascal_context.py converts pascal context from 456 categories (.mat) to 59 categories (.png -- color & gray). I have listed the mapping ids, you can also use func search_map_id to generate it.

TODO