Shape-biased Texture Agnostic Representations for Improved Textureless and Metallic Object Detection and 6D Pose Estimation

Peter Hönig, Stefan Thalhammer, Jean-Baptiste Weibel, Matthias Hirschmanner, Markus Vincze,

Datasets

• To generate the randomized texturing dataset in the BOP format use our rendering scripts provided in ./blenderproc_rendering
• to use the rendering scripts, Blenderproc is necessary: https://github.com/DLR-RM/BlenderProc.git
• copy & paste the scripts to /Blenderproc/examples/datasets/bop_challenge
• and execute e.g. with: python rerun.py run examples/datasets/bop_challenge/main_tless_random_texture.py ../datasets resources/cc_textures examples/datasets/bop_challenge/output --num_scenes=1000

Object Detection and Pose Estimation Software

We used the following list of algorithms:

• MMDetection: https://github.com/open-mmlab/mmdetection
• Pix2Pose: https://github.com/kirumang/Pix2Pose
• GDR-Net: https://github.com/THU-DA-6D-Pose-Group/GDR-Net