Skip to content

Latest commit

 

History

History
129 lines (114 loc) · 3.35 KB

README.md

File metadata and controls

129 lines (114 loc) · 3.35 KB
Raw

Prepare Datasets for TAPPS

TAPPS has builtin support for two datasets: Cityscapes Panoptic Parts (Cityscapes-PP) and Pascal Panoptic Parts (Pascal-PP) The datasets are assumed to exist in a directory specified by the environment variable DETECTRON2_DATASETS. Under this directory, detectron2 will look for datasets in the structure described below.

$DETECTRON2_DATASETS/
  cityscapes/
  pascal/

You can set the location for builtin datasets by export DETECTRON2_DATASETS=/path/to/datasets. If left unset, the default is ./datasets relative to your current working directory.

Expected dataset structure for Cityscapes-PP:

First download the Cityscapes dataset and put the data in the cityscapes directory. Download gtFine_trainvaltest.zip, leftImg8bit_trainvaltest.zip, and gtFinePanopticParts.zip. Structure it as below:

cityscapes/
  gtFine/
    train/
      aachen/
      ...
    val/
    test/
  leftImg8bit/
    train/
    val/
    test/
  gtFinePanopticParts/
    train/
    val/

In any directory, clone cityscapesScripts by:

git clone https://github.com/mcordts/cityscapesScripts.git

To create labelTrainIds.png, first prepare the above structure, then run cityscapesScripts with:

CITYSCAPES_DATASET=/path/to/abovementioned/cityscapes python cityscapesscripts/preparation/createTrainIdLabelImgs.py

To generate Cityscapes panoptic dataset, run cityscapesScripts with:

CITYSCAPES_DATASET=/path/to/abovementioned/cityscapes python cityscapesscripts/preparation/createPanopticImgs.py

To prepare the part segmentation files necessary for training, run:

python datasets/prepare_cityscapes_pp.py

After doing this, the data should be in the following structure:

cityscapes/
  gtFine/
    train/
      aachen/
      ...
    val/
    test/
    cityscapes_panoptic_train.json
    cityscapes_panoptic_train/
    cityscapes_panoptic_val.json
    cityscapes_panoptic_val/
    cityscapes_panoptic_test.json
    cityscapes_panoptic_test/
  leftImg8bit/
    train/
    val/
    test/
  gtFinePanopticParts/
    train/
    val/
  gtFineParts/
    train/
    val/
  images_val.json
  images_train.json

Expected dataset structure for Pascal-PP:

Download the Pascal-PP labels and the Pascal VOC 2010 images. Organize the data in the following structure:

pascal/
  JPEGImages/    # From VOC2010
  labels/        # From pascal_panoptic_parts_v2.0
    training/
    validation/

To generate the panoptic, semantic and part segmentation annotations and split the images into training and validation splits, run:

python datasets/prepare_pascal_pp.py

Afterwards, the data should have the following structure:

pascal/
  images/
    training/
    validation/
  labels/
    training/
    validation/
  panoptic/
    training/
    validation/
    panoptic_training.json
    panoptic_validation.json
  semantic/
    training/
    validation/
  parts/
    training/
    validation/
  images_training.json
  images_validation.json

Note: if you wish to use the Pascal-PP-107 labels (instead of the default Pascal-PP-57), also run:

python datasets/prepare_pascal_pp_107.py