dataset_download

Download Datasets

COCO 2017 Download

# Needs 1 argument, path to download directory.
bash coco-2017_download.sh ./coco

Colab Setup Only

COCO Sub Directory Split for Colab

# Since in the above COCO 2017 Download we chose ./coco
python3 sub_divide_training.py ./coco 

# Remove old train and test so you dont have to upload them too
rm -rf ./coco/train2017 ./coco/test2017

Upload to Google Drive

Install rclone

curl https://rclone.org/install.sh | sudo bash

Link rclone with google drive, figure it out yourself I believe in you :)

https://rclone.org/drive/

Upload to Google Drive

First, make a directory on your Google Drive in 'My Drive' root named COCO

# Copy to remote, remote-name is the name of the remote you created during setup
rclone copy ./coco remote-name:COCO/ -vv

# rclone copy val2017_sub-500 gdrive-usf:DATASETS/COCO/2017/val2017_sub-500 -vv --drive-chunk-size=256M --transfers=40 --checkers=40 --tpslimit=9 --fast-list --max-backlog 200000

Now wait a while :)

My subdivided COCO for Colab Google Drive Share

https://drive.google.com/drive/folders/1EVsLBRwT2njNWOrmBAhDHvvB8qrd9pXT?usp=sharing

Colab Notebook on Training Detectron2 with COCO 2017

https://colab.research.google.com/drive/1OVStblo4Q3rz49Pe9-CJcUGkkCDLcMqP