INBD

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

CVPR 2023. Paper link

---

---

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:

---

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:

• DO (Dryas octopetala)
• EH (Empetrum hermaphroditum)
• VM (Vaccinium myrtillus)

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