Converting NDDS-Data to Preprocessed Linemod

This repository can be used to easily convert the data generated with NDDS to preprocessed Linemod.

Installing the code

• Use conda to install the code

conda create -n nddsConvertor python=3.6
conda activate nddsConvertor
pip install -r requirements.txt

Using the code

This repository contains 3 different tools, which all helps you to convert your data:

1 NDDS Images Groundtruth convertor

After generating RGB, Depth, Mask and Groundtruth data with NDDS. Use the following command to convert them:

python run.py --model-number 11 ndds --ndds-path datasets/test_dataset/NDDS_generated --save-path datasets/test_dataset/generated_dataset --train-percentage 0.9 --scale 1

args are:

• --model-number: The number you want your object to be saved. In Linemod every object has a uniq number
• --ndds-path: The Path where your NDDS data is saved under.
• --save-path: The Path you want to save your generated data.

optional args are:

• --train-percentage: The percentage amount of data used for training. (1-this amount) will be used for validation
• --scale: Since there are a lot of unit conversaion from CAD to NDDS and exporting it from there. The scale is used to make the model scaled to the data. More in run.py.

2 Converted dataset viewer

After converting your data this tool can be used to see if the scale factor and the bounding boxes are correct.

python run.py --model-number 11 view --dataset-path datasets/test_dataset/generated_dataset --frame-number 1

args are:

• --model-number: The number you want your object to be saved. In Linemod every object has a uniq number
• --dataset-path: The Path to the generated dataset. optional args are:
• --frame-number: The frame that you want to visualize the annotation.
• --save-path', The Path to store visualized images

3 .ply file point extractor

For reading the points from a .ply file and make the model/models_info.yml file.

python run.py --model-number 11 ply --ply-path datasets/test_dataset/generated_dataset/models/obj_11.ply --save-path datasets/test_dataset/generated_dataset/models

Be aware that this only extracts the points from a .ply file and saves them under --save-path. models_info.yml must be created manually.

args are:

• --model-number: The number you want your object to be saved. In Linemod every object has a uniq number
• --ply-path: Path to the .ply file
• --save-path', The path to save the obj.yml containing the points

##Citation These two projects helped to understand the rotation matrices and 6DoF transformation:

• https://github.com/KatharinaSchmidt/clean-pvnet
• https://github.com/ybkscht/EfficientPose

@misc{bukschat2020efficientpose,
      title={EfficientPose: An efficient, accurate and scalable end-to-end 6D multi object pose estimation approach}, 
      author={Yannick Bukschat and Marcus Vetter},
      year={2020},
      eprint={2011.04307},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}