Fasteners

Dependencies

$ # DL VoTT snap from github
$ sudo snap install "~/Downloads/vott-2.1.0-linux.snap --dangerous
$ yay -S opencv2
$ pipenv install

CODE

• SXXX = Screw
• BXXX = Bolt
• NXXX = Nut
• OXXX = Others

Token

my_token: maRMdjSQEkuaDCcGoqmA/LSt3wotOZFxlLgrDNTlhkQ=

1. Data preparation

First of all, it's need to make directories by following cmd.

$ mkdir {datasets_raw,datasets_vott,datasets_voc}/xxx_dataset

2. VoTT Flow

do Annotion by VoTT

• Firstly, create Security Token and save it to somewhere private space
• Connections:
  • target connection
    • name is target_xxx_dataset
    • select local storage of datasets_vott/xxx_dataset
  • source connection
    • name is source_xxx_dataset
    • select local storage of datasets_raw/xxx_dataset
• Project settings
  • name is xxx_dataset
  • select security token and target/source connection that i have created
  • name is like xxx_dataset_vott.vott since there is already same name folder.
  • video frame rate is 1 sec
• Export Setings
  • providor: pascal voc
  • asset state: only tagged assets
  • test:train = 20:80
  • export unsinged: check

after annotation is done, export proj as Pascal VOC

3. Copy exported Pascal VOC files to the datasets folder

$ DSNAME=xxx_dataset;
$ cp -R ./datasets_vott/${DSNAME}/${DSNAME}-PascalVOC-export/* ./datasets_voc/${DSNAME}/

with DA

4. Auto annotation & Data augmentation

$ python scripts/augment.py --dsname=$DSNAME --size=5 --train_percentage=1

without DA

4. Auto annotation & Data augmentation

$ python scripts/voc_summary.py --dsname=$DSNAME --filename=S001.txt

5. create Darknet Data (label directory and train.txt and test.txt which have full path)

$ python scripts/voc_label.py --dsname=$DSNAME --filenames=S001.txt

REFERENCES:

• https://github.com/wakuphas/wakuphas/blob/master/AI/Scripts/increase_img.py
• https://wakuphas.hatenablog.com/entry/2018/09/19/025941
• https://github.com/zchrissirhcz/imageset-viewer
• http://mukopikmin.hatenablog.com/entry/2018/12/05/002339
• https://github.com/tzutalin/labelImg
• https://github.com/aleju/imgaug