Learning from Synthetic Dataset for Crop Seed Instance Segmentation

Overview of the proposed training process of crop seed instance segmentation.

See https://www.biorxiv.org/content/10.1101/866921v2 for details

Data included in this repository

• Codes in Jupyter Notebook format

  • Mask RCNN Inference of crop seed images

  • Multivariate Analysis and Visualization

Large Files are stored in Google Drive

https://drive.google.com/file/d/1g8bg9ter9DlKWgs0lfPZMQemRlzRVOQr/view?usp=sharing

Contents

• Barley data
  • Synthetic Images and Masks of Test Data
  • Real World Images of Test Data (19 barley cultivar)
    • The annotation of Real World Images formated in JSON
  • Trained Model Weights
• Other crops
  • Model Weights and Image of Rice seeds
  • Model Weights and Images of 4 Wheat cultivars. One model can infer 4.

Howto

1. Clone the repository

2. Install Dependencies (See below)

3. Download the data.zip from google drive and place it into the top directory of this repository

4. Run the notebook

Dependencies

• Mask RCNN implemented with Keras/Tensorflow provided by matterport.
• Keras==2.2.4
• Tensorflow-gpu==1.13.1
• pyefd==1.4.1 (for EFD analysis)
• other general packages such as sklearn, scikit-image, opencv3, etc..

Author

Yosuke Toda

JST PRESTO / ITbM, Nagoya Univ.