formatomato

Introduction

This repository provides modules for converting an image dataset to CSV filess.

Usage

1. Install Python packages by pipenv

$ pipenv install

2. Run the below command with file path arguments

$ pipenv run main.py \
--input_dataset_path=/path/to/dataset \
--dataset_name=<dataset_name>

Arguments

Argument	Explanation
--config_path /path/to/config.yaml	A path to an yaml file for configuration instead of giving the following arguments
--input_dataset_path /path/to/dataset	A path to an input dataset
--dataset_name {coco, imagenet}	A path to the input dataset
--output_csv_path /path/to/annotations.csv	(optional) A path to an output CSV file for annotations
--output_image_csv_path /path/to/image.csv	(optional) A path to an output CSV file for images
--output_classes_csv_path	(optional) A path to an output CSV file for classes
--normalize	(optional) A flag for normalizing coordinate values of bbox

Input dataset

Check --input_dataset_path for given --dataset_name

--dataset_name coco

Give --input_dataset_path /path/to/coco2017/annotations_trainval2017/annotations/instances_val2017.json

--dataset_name imagenet

Give --input_dataset_path /path/to/ILSVRC2012/ILSVRC2012_img_train/train

/path/to/ILSVRC2012/ILSVRC2012_img_train/train
    - n01440764
        - n01440764_10026.JPEG
        - n01440764_10027.JPEG
        ...
    - n01443537
        - n01443537_10007.JPEG
        - n01443537_10014.JPEG
    ...

Output

The following 3 files:

File	Explanation
out.csv	Annotations
out_image.csv	Image infos
out_classes	Defined classes

Each file consists of the following columns.

out.csv

Column	Explanation
IMAGE_ID	Image IDs
IMAGE_NAME	Image names
CLASS	Annotated class names denoting labels for each images in a classification dataset or for each bbox in a detection dataset
X_MIN, Y_MIN	(detection) Upper-left coordinates for each bbox
X_MAX, Y_MAX	(detection) Lower-right coordinates for each bbox
WIDTH, HEIGHT	(detection) Widths and heights for each bbox

out_image.csv

Column	Explanation
IMAGE_ID	Image IDs
IMAGE_NAME	Image names
WIDTH, HEIGHT	Widths and heights for each images

out_classes

Column	Explanation
CLASS_ID	Class IDs
SUPERCLASS	Superclass names
CLASS	Class names