Skip to content

Latest commit

 

History

History
223 lines (187 loc) · 7.52 KB

README.md

File metadata and controls

223 lines (187 loc) · 7.52 KB
Raw

Prepare Datasets for ODISE

Dataset preparation for ODISE follows Detectron2 and Mask2Former.

A dataset can be used by accessing DatasetCatalog for its data, or MetadataCatalog for its metadata (class names, etc). This document explains how to setup the builtin datasets so they can be used by the above APIs. Use Custom Datasets gives a deeper dive on how to use DatasetCatalog and MetadataCatalog, and how to add new datasets to them.

ODISE has builtin support for a few datasets. 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, if needed.

$DETECTRON2_DATASETS/
  ade/
  coco/
  VOCdevkit/
  pascal_ctx_d2/
  pascal_voc_d2/

You can set the location for builtin datasets by export DETECTRON2_DATASETS=/path/to/datasets. If left unset, the default is ./datasets under the ODISE project directory.

The model zoo contains configs and models that use these builtin datasets.

Expected dataset structure for COCO:

coco/
  annotations/
    instances_{train,val}2017.json
    panoptic_{train,val}2017.json
    captions_{train,val}2017.json
    # below are prepare_coco_caption.py
    panoptic_caption_{train,val}2017.json  
  {train,val}2017/
  panoptic_{train,val}2017/ 
  # below are generated by prepare_coco_semantic_annos_from_panoptic_annos.py
  panoptic_semseg_{train,val}2017/  #

Download the dataset from http://cocodataset.org/#download:

cd $DETECTRON2_DATASETS
wget http://images.cocodataset.org/zips/train2017.zip
unzip train2017.zip -d coco/
wget http://images.cocodataset.org/zips/val2017.zip 
unzip val2017.zip -d coco/
wget http://images.cocodataset.org/annotations/annotations_trainval2017.zip
unzip annotations_trainval2017.zip -d coco/
wget http://images.cocodataset.org/annotations/panoptic_annotations_trainval2017.zip
unzip panoptic_annotations_trainval2017.zip -d coco/
unzip coco/annotations/panoptic_train2017.zip -d coco/
unzip coco/annotations/panoptic_val2017.zip -d coco/

Install the panopticapi (also automatically done by installing ODISE) by:

pip install git+https://github.com/cocodataset/panopticapi.git

Generate the semantic segmentation annotations coco/panoptic_semseg_{train,val}2017/ by running:

python datasets/prepare_coco_semantic_annos_from_panoptic_annos.py

to extract the semantic annotations from the panoptic ones (only used for evaluation).

Generate the panoptic annotations with COCO captions panoptic_caption_{train,val}2017.json by running:

python datasets/prepare_coco_caption.py

Expected dataset structure for ADE20k (A-150) and ADE20k-Full (A-847):

ade/
  ADEChallengeData2016/
    images/
    annotations/
    objectInfo150.txt
    # downloaded instance annotation
    annotations_instance/
    # generated by prepare_ade20k_sem_seg.py
    annotations_detectron2/
    # generated by prepare_ade20k_ins_seg.py
    ade20k_instance_{train,val}.json
    # generated by prepare_ade20k_pan_seg.py
    ade20k_panoptic_{train,val}.json
    ade20k_panoptic_{train,val}/
    
  ADE20K_2021_17_01/
    images/
    index_ade20k.pkl
    objects.txt
    # generated by prepare_ade20k_full_sem_seg.py
    images_detectron2/
    annotations_detectron2/

ADE20k(A-150)

Download the dataset from http://sceneparsing.csail.mit.edu/:

cd $DETECTRON2_DATASETS
wget http://data.csail.mit.edu/places/ADEchallenge/ADEChallengeData2016.zip
# generate folder ade/ADEChallengeData2016/
unzip ADEChallengeData2016.zip -d ade/

Download the instance annotations from http://sceneparsing.csail.mit.edu/:

cd $DETECTRON2_DATASETS
wget http://sceneparsing.csail.mit.edu/data/ChallengeData2017/annotations_instance.tar
# generate folder ade/ADEChallengeData2016/annotations_instance/
tar -xvf annotations_instance.tar -C ade/ADEChallengeData2016/

Generate the directory ade/ADEChallengeData2016/annotations_detectron2 by running:

python datasets/prepare_ade20k_sem_seg.py

Generate instance annotations ade/ADEChallengeData2016/ade20k_instance_{train,val}.json by running:

python datasets/prepare_ade20k_ins_seg.py

Generate panoptic annotations ade/ADEChallengeData2016/ade20k_panoptic_{train,val}.json and ade/ADEChallengeData2016/ade20k_panoptic_{train,val} by running:

python datasets/prepare_ade20k_pan_seg.py

ADE20k-Full(A-847)

Register and download the dataset from https://groups.csail.mit.edu/vision/datasets/ADE20K/:

cd $DETECTRON2_DATASETS
wget your/personal/download/link/{username}_{hash}.zip
unzip {username}_{hash}.zip -d ade/

Generate the directories ade/ADE20K_2021_17_01/images_detectron2 and ade/ADE20K_2021_17_01/annotations_detectron2 by running:

python datasets/prepare_ade20k_full_sem_seg.py

Expected dataset structure for PASCAL Context (PC-59), PASCAL Context Full (PC-459) and PASCAL VOC (PAS-21):

VOCdevkit/
  VOC2012/
    Annotations/
    JPEGImages/
    ImageSets/
      Segmentation/
  VOC2010/
    JPEGImages/
    trainval/
    trainval_merged.json
# generated by prepare_pascal_voc_sem_seg.py
pascal_voc_d2/
  images/
  annotations_pascal21/
# generated by prepare_pascal_ctx_sem_seg.py
pascal_ctx_d2/
  images/
  annotations_ctx59/
  # generated by prepare_pascal_ctx_full_sem_seg.py
  annotations_ctx459/

PASCAL VOC (PAS-21)

Download the dataset from http://host.robots.ox.ac.uk/pascal/VOC/:

cd $DETECTRON2_DATASETS
wget http://host.robots.ox.ac.uk/pascal/VOC/voc2012/VOCtrainval_11-May-2012.tar
# generate folder VOCdevkit/VOC2012
tar -xvf VOCtrainval_11-May-2012.tar

Generate directory pascal_voc_d2 running:

python datasets/prepare_pascal_voc_sem_seg.py

PASCAL Context (PC-59)

Download the dataset from http://host.robots.ox.ac.uk/pascal/VOC/ and annotation from https://www.cs.stanford.edu/~roozbeh/pascal-context/:

cd $DETECTRON2_DATASETS
wget http://host.robots.ox.ac.uk/pascal/VOC/voc2010/VOCtrainval_03-May-2010.tar
# generate folder VOCdevkit/VOC2010
tar -xvf VOCtrainval_03-May-2010.tar 
wget https://www.cs.stanford.edu/~roozbeh/pascal-context/trainval.tar.gz
# generate folder VOCdevkit/VOC2010/trainval
tar -xvzf trainval.tar.gz -C VOCdevkit/VOC2010 
wget https://codalabuser.blob.core.windows.net/public/trainval_merged.json -P VOCdevkit/VOC2010/

Install Detail API by:

git clone https://github.com/zhanghang1989/detail-api.git
rm detail-api/PythonAPI/detail/_mask.c
pip install -e detail-api/PythonAPI/

Generate directory pascal_ctx_d2/images and pascal_ctx_d2/annotations_ctx59 running:

python datasets/prepare_pascal_ctx_sem_seg.py

PASCAL Context Full (PC-459)

Generate directory pascal_ctx_d2/annotations_ctx459 running:

python datasets/prepare_pascal_ctx_full_sem_seg.py