Updated on Nov. 7th, 2017.

---

KOMATSUNA dataset for instance segmentation, tracking and reconstruction
We provide two datasets for 3D plant phenotyping such as leaf segmentation, tracking and reconstruction. Multiview dataset contains high resolution images of plants captured with three fixed cameras, and images for geometric and color calibration. RGBD dataset contains both RGB and depth images of plants captured with a RGB-D camera. For both datasets, we made plant images by cropping an individual plant in original images, and label images by manually drawing ground truth regions of leaves in plant images. The detail of each dataset is described in the readme file. Note that plants in each dataset are different.

• Multi-view dataset
  
• readme
• original images
• plant images
• label images
• images for calibration
• RGB-D dataset
• readme
• original images
• aligned depth images (original)
  
• aligned depth images (ours)
  
• plant images
• label images
• Annotation tool
  
• Instance divider ver1.00


Reference

Hideaki Uchiyama, Shunsuke Sakurai, Masashi Mishima, Daisaku Arita, Takashi Okayasu, Atsushi Shimada and Rin-ichiro Taniguchi, "An easy-to-setup 3D phenotyping platform for KOMATSUNA dataset," ICCV Workshop on Computer Vision Problems in Plant Phenotyping, pp.2038-2045, 2017. [link]

ACKNOWLEDGEMENT
We thank Masashi Mishima and Shunsuke Sakurai for capturing and cropping part of the dataset.


---

Contact: