Skip to content

Latest commit

 

History

History
105 lines (65 loc) · 3 KB

README.md

File metadata and controls

105 lines (65 loc) · 3 KB
Raw

INBD

Iterative Next Boundary Detection for Instance Segmentation of Tree Rings in Microscopy Images of Shrub Cross Sections

CVPR 2023. Paper link

Example input image and detected tree rings

Setup:

Python version: 3.7. Other versions are known to cause issues.

#setup virtualenv
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

#download dataset
python fetch_dataset.py

#download pretrained models
python fetch_pretrained_models.py

Or use GitHub Codespaces: Open in GitHub Codespaces

Inference

#single imagefile
python main.py inference checkpoints/INBD_EH/model.pt.zip dataset/EH/inputimages/EH_0033.jpg

#list of imagefiles
python main.py inference checkpoints/INBD_EH/model.pt.zip dataset/EH/test_inputimages.txt

Training:

#first, train the 3-class segmentation model
python main.py train segmentation           \
  dataset/EH/train_inputimages.txt          \
  dataset/EH/train_annotations.txt

#next, train the inbd network
python main.py train INBD \
  dataset/EH/train_inputimages.txt          \
  dataset/EH/train_annotations.txt          \
  --segmentationmodel=checkpoints/segmentationmodel/model.pt.zip   #adjust path

Dataset

We introduce a new publicly available dataset: MiSCS (Microscopic Shrub Cross Sections)

The dataset and annotations can be downloaded via python fetch_dataset.py or via the following links:

All images were acquired by Alba Anadon-Rosell. If you have ecology-related questions, please contact a.anadon at creaf.uab.cat

If you want to use this dataset for computer vision research, please cite the publication as below.

Citation

@inproceedings{INBD,
  title     = "{Iterative Next Boundary Detection for Instance Segmentation of Tree Rings in Microscopy Images of Shrub Cross Sections}",
  author    = {Alexander Gillert and Giulia Resente and Alba Anadon‐Rosell and Martin Wilmking and Uwe von Lukas},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  month     = {June},
  year      = {2023},
  pages     = {14540-14548}
}

License

License for the source code: MPL-2.0

License for the dataset: CC BY-NC-SA 4.0