1. Multi-Level Feature Re-weighted fusion for the Semantic Segementation of crops and weeds

A proposed network for the pixel-wise semantic segmentation of crops and weed with minimal memory overhead, it is experimented on three commonly used datasets

2. Network Training and Testing

It was trained and tested using:

• NVIDIA Tesla P40 GPUs
• PyTorch 1.11.0

3. Dataset and the Experimental results

The datasets for testing and the model files must in the project root directory, which can be access via the following links to google drive

• Test datasets and model files

Running commands for the Test

The experiment can be tested using the following commands, where the dataset parameter can be changed for the different datasets bweeds : Bonirob , cweeds : CWFID, rweeds : Rice seedlings

• Baseline (bweeds, cweeds, rweeds)
• cmd: python main_ours.py --dataset='bweeds' --backbone='baseline'

• MFF (bweeds, cweeds, rweeds)
• cmd: python main_ours_nostream.py --dataset='bweeds' --backbone='ours_l34rw_partial_weight'

• MFRWF (bweeds, cweeds, rweeds)
• cmd: python main_ours_nostream.py --dataset='bweeds' --backbone='ours_l34rw_partial_decoder'

• MFRWF + CWF (bweeds, cweeds, rweeds)
• cmd: python main_ours_nostream.py --dataset='bweeds' --backbone='ours_l34rw_fully'

---

---

Visulaization of the Results

• Bonirob:
  
• CWFID:
  
• Rice Seedlings: