COCO to Yolo Downloader (Images, Annoatations, Negatives)

• Downloads COCO dataset by multiple image categories in parallel threads, converts COCO annotations to YOLO format and stored in respective .txt files
• Download Negative images which excludes the categories in categories_to_download.txt
• Can include custom class numbers to be added to annoation, just add desired numbers to categories_to_download.txt
• Can limit downloads based on per class

Requirements:

• pip3 install pycococtools
• COCO Dataset "instances_train2017.json" file from here

Run:

• Update categories_to_download.txt file with category names and calss numbers that you want to add.
• To download coco dataset with limit of 5000 images per class:

 python3 main.py -o download -l 5000

• To download negatives from coco dataset with limit of 1000:

python3 main.py -o negatives -l 1000

• By default annotations along with images are stored in annotations folder, you can change it here
• Image download will be skipped if already exists. However, annotations will always be appended to the existing file so it's advisable to delete *.txt file before a re-run
• Do not delete or modify coco-names.txt