MSCOCO

The official MSCOCO API is hard to use for me.
This script creates images for semantic/instance segmentation, object detection and image caption.

how to use

step1.

clone the official API

git clone https://github.com/rkuga/MSCOCO.git
cd MSCOCO
git clone https://github.com/pdollar/coco.git

step2.

make init.py

vim coco/__init__.py
vim coco/PythonAPI/__init__.py

step3.

download the dataset
your data_dir is
yourpath|---annotations
　　　　　　　|---train2014
　　　　　　　|---val2014

step4.

for instance segmentations

python coco.py --data_dir yourpath --data_type train --mode instances

for semantic segmentations

python coco.py --data_dir yourpath --data_type train --mode semantics

with categorical images

for bounding boxs

python coco.py --data_dir yourpath --data_type train --mode bboxs

Bbox's color means its object class

for captions

python coco.py --data_dir yourpath --data_type train --mode captions

instances mode takes so much time.