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

• [x] Converting scripts for PASCAL Context dataset
• [ ] Scripts for ADE20k