SeedSense

• This repository uses SFA-Net based image-segmentation architecture to determine areas in which we can sow seeds autonomously.
• Try this project for yourself: Working demo

Install

conda create -n ssenv python=3.11
conda activate ssenc
pip install torch==2.6.0 torchvision==0.21.0 torchaudio==2.6.0
pip install -r requirements.txt

Folder Structure

Prepare the following folders to organize this repo:

SFA-Net
├── network
├── config
├── data_pred
├── tools
├── model_weights (save the model weights)
├── fig_results (save the masks predicted)
├── lightning_logs (CSV format training logs)
├── data
│   ├── LoveDA
│   │   ├── Train
│   │   │   ├── Urban
│   │   │   │   ├── images_png (original images)
│   │   │   │   ├── masks_png (original masks)
│   │   │   │   ├── masks_png_convert (converted masks used for training)
│   │   │   │   ├── masks_png_convert_rgb (original rgb format masks)
│   │   │   ├── Rural
│   │   │   │   ├── images_png 
│   │   │   │   ├── masks_png 
│   │   │   │   ├── masks_png_convert
│   │   │   │   ├── masks_png_convert_rgb
│   │   ├── Val (the same with Train)
│   │   ├── Test
│   │   ├── train_val (Merge Train and Val)

---

Note: To access pre-trained models, you can download them from LoveDA weights. Place the downloaded files in the model_weights directory.

---

Data Preprocessing

Download Dataset: LoveDA

Configure the folder as shown in 'Folder Structure' above.

python tools/loveda_mask_convert.py --mask-dir data/LoveDA/Train/Rural/masks_png --output-mask-dir data/LoveDA/Train/Rural/masks_png_convert

python tools/loveda_mask_convert.py --mask-dir data/LoveDA/Train/Urban/masks_png --output-mask-dir data/LoveDA/Train/Urban/masks_png_convert

python tools/loveda_mask_convert.py --mask-dir data/LoveDA/Val/Rural/masks_png --output-mask-dir data/LoveDA/Val/Rural/masks_png_convert

python tools/loveda_mask_convert.py --mask-dir data/LoveDA/Val/Urban/masks_png --output-mask-dir data/LoveDA/Val/Urban/masks_png_convert

This alters the LoveDa dataset images in training set to improve the water detection capability of the model

python tools/enhance_loveda.py

Training

"-c" means the path of the config

python train.py -c config/uavid/sfanet.py

Testing

"-c" denotes the path of the config, Use different config to test different models.

"-o" denotes the output path

"-t" denotes the test time augmentation (TTA), can be [None, 'lr', 'd4'], default is None, 'lr' is flip TTA, 'd4' is multiscale TTA

"--rgb" denotes whether to output masks in RGB format

LoveDA (Online Testing)

• To get RGB files:

python test_loveda.py -c config/loveda/sfanet.py -o fig_results/loveda/sfanet_loveda --rgb -t "d4"

• For submitting to the online test site:

python test_loveda.py -c config/loveda/sfanet.py -o fig_results/loveda/sfanet_loveda -t "d4"

• For generating the prediction for one image:

python prediction.py -c config/loveda/sfanet.py -i /path/to/image -o /path/to/output -t "d4" 

Using the Streamlit App

The Streamlit app serves as an interactive interface to run the plantable area detection pipeline. Follow these steps to use the app:

1. Launch the App
   Open a terminal in the repository's root directory and run:

   streamlit run app.py

   Streamlit will start and open your default browser to display the app interface.

2. Use the Interface

   • Input Files: Use the sidebar and file uploader components to select image files (PNG, JPG, JPEG) and a metadata CSV file.
   • Configuration: Adjust parameters like hex grid spacing, device (cuda/cpu), and model type directly in the sidebar.
   • Run Pipeline: Once your files are uploaded and validated, click the "🚀 Run Pipeline" button. The pipeline will process the images, update the metadata, and output results.
   • View Results: After processing, the app will display outputs such as the stitched mask with a hex grid overlay and the calculated green area percentage.

Acknowledgement

• SFA-Net
• pytorch lightning
• timm
• pytorch-toolbelt
• ttach
• catalyst
• mmsegmentation
• GeoSeg