Skip to content

Latest commit

 

History

History
109 lines (88 loc) · 3.69 KB

DATA.md

File metadata and controls

109 lines (88 loc) · 3.69 KB
Raw

Dataset preparation

If you want to reproduce the results in the paper for benchmark evaluation and training, you will need to setup dataset.

COCO

  • Download the images (2017 Train, 2017 Val, 2017 Test) from coco website.

  • Download annotation files (2017 train/val and test image info) from coco website.

  • Place the data (or create symlinks) to make the data folder like:

    ${CenterNet_ROOT}
|-- data
`-- |-- coco
    `-- |-- annotations
        |   |-- instances_train2017.json
        |   |-- instances_val2017.json
        |   |-- person_keypoints_train2017.json
        |   |-- person_keypoints_val2017.json
        |   |-- image_info_test-dev2017.json
        `-- images
            |-- train2017
            |-- val2017
            |-- test2017

  • [Optional] If you want to train ExtremeNet, generate extreme point annotation from segmentation:

    cd $CenterNet_ROOT/tools/
python gen_coco_extreme_points.py

    It generates instances_extreme_train2017.json and instances_extreme_val2017.json in data/coco/annotations/.

Pascal VOC

  • Run

    cd $CenterNet_ROOT/tools/
bash download_pascal_voc.sh

  • The above script includes:

    • Download, unzip, and move Pascal VOC images from the VOC website.
    • Download Pascal VOC annotation in COCO format (from Detectron).
    • Combine train/val 2007/2012 annotation files into a single json.

  • Move the created voc folder to data (or create symlinks) to make the data folder like:

    ${CenterNet_ROOT}
|-- data
`-- |-- voc
    `-- |-- annotations
        |   |-- pascal_trainval0712.json
        |   |-- pascal_test2017.json
        |-- images
        |   |-- 000001.jpg
        |   ......
        `-- VOCdevkit

    The VOCdevkit folder is needed to run the evaluation script from faster rcnn.

KITTI

  • Download images, annotations, and calibrations from KITTI website and unzip.

  • Download the train-val split of 3DOP and SubCNN and place the data as below

    ${CenterNet_ROOT}
|-- data
`-- |-- kitti
    `-- |-- training
        |   |-- image_2
        |   |-- label_2
        |   |-- calib
        |-- ImageSets_3dop
        |   |-- test.txt
        |   |-- train.txt
        |   |-- val.txt
        |   |-- trainval.txt
        `-- ImageSets_subcnn
            |-- test.txt
            |-- train.txt
            |-- val.txt
            |-- trainval.txt

  • Run python convert_kitti_to_coco.py in tools to convert the annotation into COCO format. You can set DEBUG=True in line 5 to visualize the annotation.

  • The data structure should look like:

    ${CenterNet_ROOT}
|-- data
`-- |-- kitti
    `-- |-- annotations
        |   |-- kitti_3dop_train.json
        |   |-- kitti_3dop_val.json
        |   |-- kitti_subcnn_train.json
        |   |-- kitti_subcnn_val.json
        `-- images
            |-- trainval
            |-- test