Pflückroboter

Data-Synthesizer

CLI usage

usage: main.py synth [-h] [-f [{tomato,bell_pepper,cucumber,strawberry} ...]] [-m {verification,coco}] [--debug] [--pos POS] [--no NO] [--out OUT]

options:
  -h, --help            show this help message and exit
  -f [{tomato,bell_pepper,cucumber,strawberry} ...]
  -m {verification,coco}
  --debug               launches blenderproc debug environment
  --no NO               rendering quantity
  --out OUT             name of output folder

camera:
  --pos POS             camera position quantity per render

Detailed

Required

-f (Fruit)

• Multiple fruits can be chosen
• Same fruit can be chosen multiple times ("-f tomato tomato" is valid)
• List will be cycled until amount of renderings specified is done

-m (Mode)

• Only one mode supported at a time
• verification produces .hdf5 images with a .yaml with validation data
• coco produces .jpg images with a .json with coco annotations data

Optional

--debug (Debug-mode)

• starts the data-synthesizer in debug mode, launching in blender
• all other arguments are ignored
• arguments for testing in debug have to be set in the debug script, which opens automatically

--no (Rendering Quantity)

• Sets the amount of Configurations that are to be generated
• Defaults to 1

--out (Output folder)

• Sets the Output folder for the data
• Default is /data_synthesizer/output/verification/ and /data_synthesizer/output/coco/ respectively
• Without this option set, existing data will never be overwritten/appended
• With this option set, new data will be appended to existing data

--pos (Camera positions)

• Sets the amount of camera positions per configuration/rendering
• Defaults to 1
• Positions will be evenly placed in a half circle around the plant

Examples

python main.py -f tomato bell_pepper -m coco --no 2

The output folder will be /data_synthesizer/coco_xxx/, with xxx being the lowest unused number. The 2 .jpg images will be placed into /images/ in this folder, they show one plant with tomatos, one with bell peppers. The COCO-annotations will be saved in the coco_annotations.json in the coco folder.

python main.py -f tomato bell_pepper -m coco 

This is valid code, but it will only produce a single image with a tomato

python main.py -f tomato bell_pepper -m verification --no 10 --pos 5

This will output to the default folder, as described above, but replacing coco with verification. Every 5 images in a row are of the same fruit type and show the same plant. In total, this will produce 50 images and one .yaml file.

python main.py --debug

Opens blender with the data-synthesizer script open in the blender code editor. The following code can be found at the bottom of the file:

 if args.debug:
        fruit: str = "tomato"
        pos: int = 5
        tb = Testbench()
        tb.generate(fruit=fruit, pos=pos)

The variables "fruit" and "pos" can be changed to debug different settings. To run the script in blender, click on the button "Run Blenderproc", at the top of the code editor.

Pose-Estimator

python main.py preprocess path/to/dir

Preprocess the training data

python main.py train 

start training

python main.py benchmark path/to/valitaion_data.yaml

benchmark pose estimator