Skip to content

Latest commit

 

History

History
168 lines (120 loc) · 4.89 KB

README.md

File metadata and controls

168 lines (120 loc) · 4.89 KB
Raw

Tools

This directory includes tools which might be helpful for working with the CPPE-5 dataset. We also include easy to use and examples of running each tool to help you easily get started. Finally, these tools are not only useful for this dataset but can be used with other datasets as well.

Note: In each of the examples in this document, you would be expected to run the command from the respository root and not from the tools directory.

Download Dataset

The download_data.sh is a script to easily download, extract and maintain a consistent directory structure while downloading the dataset. Though you would be aple to replicate results following your own directory structure, we recommend using this script or the Python package to download the data.

Usage

Run the following command to run the script:

bash tools/download_data.sh

You can also use the Python package to download the data

  • You should first download the Python package:
pip install cppe5
  • You are now ready to download the data:
import cppe5

cppe5.download_data()

Download TF Records

The download_tfrecords.sh is a script to easily download, extract and maintain a consistent directory structure while downloading the TF Record files. Though you would be aple to replicate results following your own directory structure, we recommend using this script or the Python package to download the data.

Usage

Run the following command to run the script:

bash tools/download_tfreocrds.sh

You can also use the Python package to download the data

  • You should first download the Python package:
pip install cppe5
  • You are now ready to download the data:
import cppe5

cppe5.download_tfrecords()

Convert the PNG images to JPG images

The png2jpg.py script is a Python script to convert the PNG images in the dataset to JPG images while also converting the annotation files.

Note: This script is intended only for COCO style annotations.

Usage

usage: png2jpg.py [-h] [--default [DEFAULT]] [--png_dir PNG_DIR] [--jpg_dir JPG_DIR] [--num_images NUM_IMAGES] [--annotation_file ANNOTATION_FILE]

optional arguments:
  -h, --help            show this help message and exit
  --default [DEFAULT]   Use the default setting and paths to convert png to jpg
  --png_dir PNG_DIR     Path to the directory containing png images
  --jpg_dir JPG_DIR     Path to the directory to save jpg images
  --num_images NUM_IMAGES
                        Number of images to convert
  --annotation_file ANNOTATION_FILE
                        Path to the annotation file

Examples

If you downloaded data from the download_data.sh script above, you can directly run the following command to convert the PNG images to JPG images and update the annotations:

python tools/png2jpg.py --default

If you follow a different directory struccture you should use the following command, changing the arguments according to your directory structure:

python tools/png2jpg.py \
    --png_dir data/images \
    --jpg_dir data/images \
    --annotation_file data/annotations/train.json \
    --num_images 100

Convert Pascal VOC format to COCO

The voc2coco.py contains the script to convert the Pascal VOC XML format to COCO JSON. The dataset annotation XMLs should be stored under annotations directory and the images in the images directory. The test_ids.txt (or a text file for any other split) containing a sequence of images names to be included in the split without the extension.

Usage

usage: voc2coco.py [-h] [--ann_dir ANN_DIR] [--ann_ids ANN_IDS] [--ann_paths_list ANN_PATHS_LIST] [--labels LABELS] [--output OUTPUT] [--ext EXT]

This script support converting voc format xmls to coco format json

optional arguments:
  -h, --help            show this help message and exit
  --ann_dir ANN_DIR     path to annotation files directory. It is not need when use --ann_paths_list
  --ann_ids ANN_IDS     path to annotation files ids list. It is not need when use --ann_paths_list
  --ann_paths_list ANN_PATHS_LIST
                        path of annotation paths list. It is not need when use --ann_dir and --ann_ids
  --labels LABELS       path to label list.
  --output OUTPUT       path to output json file
  --ext EXT             additional extension of annotation file

The below command is an example to run the converter:

python tools/voc2coco.py \
    --ann_dir data/annotations/ \
    --output data/annotations/test.json \
    --ann_ids test_ids.txt \
    --labels labels.txt \
    --ext xml

Correct COCO Dataset

The coco_corrector.py contains the script to correct the COCO dataset to use relative image paths. This should not be required now with the final release of the dataset

Usage

The below command runs this script:

python tools/coco_corrector.py