Overview of the proposed training process of crop seed instance segmentation.
See https://www.biorxiv.org/content/10.1101/866921v2 for details
-
Codes in Jupyter Notebook format
https://drive.google.com/file/d/1g8bg9ter9DlKWgs0lfPZMQemRlzRVOQr/view?usp=sharing
- 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.
-
Clone the repository
-
Install Dependencies (See below)
-
Download the data.zip from google drive and place it into the top directory of this repository
-
Run the notebook
- 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..
Yosuke Toda
JST PRESTO / ITbM, Nagoya Univ.