- faCRSA
An automated pipeline for high-throughput analysis of crop root system architecture.
โค๏ธ Try faCRSA at https://facrsa.aiphenomics.com/. You can see help documents about faCRSA usage on the Github page (Section 3. Usage).
- faCRSA has been tested under Ubuntu 18.04 LTS, CentOS 7, macOS 12 Monterey, and Windows 10 with Python 3.6.0.
- Install with Conda or manually.
โ ๏ธ Before installing faCRSA, you must install Anaconda.
# Create a clear environment for faCRSA
conda create -n facrsa python=3.6
conda activate facrsa
# Install faCRSA
pip install faCRSA
# Launch task queue
facrsa-queue
# Create a new cmd window
# Launch faCRSA web
conda activate facrsa
facrsa-web
Copy the URL address (e.g. http://127.0.0.1:5000/) output from the cmd window and open it in the browser.
# We recommend cloning the faCRSA repository into a clear folder.
cd {your folder}
git clone https://github.com/njauzrn/faCRSA.git
cd faCRSA
# Create a clear environment for faCRSA
conda create -n facrsa python=3.6
conda activate facrsa
# Install Python requirements
pip install -r requirements.txt
# Launch task queue
python huey_consumer.py task_queue.huey
# Create a new cmd window
# Launch faCRSA web
conda activate facrsa
flask run
Copy the URL address (e.g. http://127.0.0.1:5000/) output from the cmd window and open it in the browser.
When you visit the web page for the first time, it will automatically jump to the initialization page. You can set SMTP server information in this page, which used to notify task status. If you don't need this function, please click the following link to skip.
- Click "Submit a new task" to load a task. We recommend you create an account to manage tasks and upload private plugins.
- Input task information and upload images.
- Task Name: set a name to distinguish different tasks.
- Description: set some information about this task.
- Pixel-to-cm Conversion Factor (CF):
- Segmentation Plugin: select a model to segment root images (default: RootSeg). You can upload a private plugin to segment more types of crops and imaging backgrounds (click here to study how to develop it).
- E-mail (Optional): receive task notification emails.
- Upload:
- Only several formats (
jpg / png / zip
) are allowed to upload. - A single task can only upload the same format.
- Only several formats (
- Submit your task and it will automatically redirect to the schedule page.
- You will be redirected to the result page and receive a task notification email.
- When the task is completed, you will receive a notification email with a link to the result page.
- You can viste each results online or download them as a zipfile.
- Click "Add a plugin" in the "My Plugin" page.
- Input plugin information and upload files.
Select each plugin in the "add task" page.
๐ Please click https://file.aiphenomics.com/demo_plugin.zip to download the demo code of private plugin.
numpy==1.19.2
tensorflow==2.4.0
The deep learning model must be constructed by Kears in Tensorflow (code: from tensorflow.keras.layers import *
).
- Plugin structure:
- network.py: the deep learning model constructed in the Python programming language.
This file must include a function named "main" without any parameters.
- weight.h5: model weight file (must be .h5 format)
- Package these files in zip format.
- weight.h5: model weight file (must be .h5 format)
- Package these files in zip format.
- Test the availability of each plugin and upload it to faCRSA.
RootSeg could be trained for images from other crops and imaging backgrounds.
- Download the model folder (
faCRSA/RootSeg
) and unzip the package. - Use the produce_train_txt script to generate a training file that includes image paths.
- Open the train script and set customized parameters, such as learning rate, training epoch and date folder.
- Switch to the training environment (e.g. facrsa) and start model training.
-
2022.08.31
- Fix abnormal program operation caused by file name.
- Fix task failure causing queue termination.
- Add an update notification on the home page.
-
2022.08.22
- Merge a horizontally flipped image with the initial masked image to improve segment accuracy.
-
2022.08.21
- Add the model training scripts for users to easily train the model.
-
2022.08.06
- Support to measure root angle (
Alpha
Version).
-
Fix several bugs.
-
Increase the speed of analysis.
- Support to measure root angle (
-
Update weight files.
-
Improve the measurement accuracy of root angle.
If you have any questions or suggestions, please contact Ruinan Zhang ([email protected]) at Nanjing Agricultural University.