[Dataset] COCO dataset to YOLO

COCO dataset

https://cocodataset.org/#download

Download list
Image : Train, Val 2017
Annotation : Train/val 2017

COCO API

install pycocotools

$ pip install pycocotools

Get User interested category images with COCO api

from pycocotools.coco import COCO
import requests
import shutil
import os

# instantiate COCO specifying the annotations json path
coco = COCO('annotations/instances_train2017.json')
coco_val = COCO('annotations/instances_val2017.json')

# Specify a list of category names of interest
# coco_class = [
#     #'person',
#     'bicycle',
#     'car',
#     'motorcylce',
#     'bus',
#     'truck']

# put class name to download
class_name = ['truck']

catIds = coco.getCatIds(catNms=class_name)
# Get the corresponding image ids and images using loadImgs
imgIds = coco.getImgIds(catIds=catIds)
images = coco.loadImgs(imgIds)

catIds_val = coco_val.getCatIds(catNms=class_name)
# Get the corresponding image ids and images using loadImgs
imgIds_val = coco_val.getImgIds(catIds=catIds_val)
images_val = coco_val.loadImgs(imgIds_val)
im_len = len(images)
for im in images:
    #if you already download whole dataset (copy)
    shutil.copy('images/train2017/' + im['file_name'], 'images/esens_coco_train/')
    #if you did not download dataset (download)
    # img_data = requests.get(im['coco_url']).content
    # with open('images/esens_coco_val/' + im['file_name'], 'wb') as handler:
    #     handler.write(img_data)
print(len(images),"train images copy complete")

# Save the images into a local folder
for im in images_val:
    shutil.copy('images/val2017/' + im['file_name'], 'images/esens_coco_val/')
    # img_data = requests.get(im['coco_url']).content
    # with open('images/esens_coco_val/' + im['file_name'], 'wb') as handler:
    #     handler.write(img_data)
print(len(images_val),"validation images copy complete")

COCO-manager

filter annotation file to user defined classes

https://github.com/immersive-limit/coco-manager

Convert2YOLO

convert COCO json format annotation file to txt format that YOLO uses

https://github.com/ssaru/convert2Yolo