Preparation

1. Please see dataset/README.md to get more details about our datasets-VIL100

2. Please see INSTALL.md to install environment and evaluation tools

3. Before training, we should download datasets-VIL100 and models

   (Google Drive) : https://drive.google.com/drive/folders/178_SSeQ4M1hI3BrTonhiTrpOWTEAenLE

   • datasets (Baidu Disk) : https://pan.baidu.com/s/1hFPKt4az6AiMmsV4c9Odmg - yyl7

   • models (Baidu Disk) : https://pan.baidu.com/s/19saFDFb6eqt3syIlnD2u2g - 2i9o

     • Trained model on the first training stage: initial_STM
     • Trained model on the second training stage: resume STM and resume ATT

4. Put them under this structure

     MMA-Net
          |----INSTALL.md
          |----README.md
          |----dataset
          |------|-----VIL100
          |----models
          |----evaluation
          |----options.py
          |----libs
          |----requirements.txt
          |----train.py
          |----test.py
   

Training and Testing

1. To train the MMA network, run following command

   python3 train.py --gpu ${GPU-IDS}

2. To test the MMA network, run following command

   python3 test.py

   The test results will be saved as indexed png file at ${root}/${output}/${valset}.

   Additionally, you can modify some setting parameters in options.py to change training configuration.

Evaluation

1. generate accuracy, fp, fp

   python evaluate_acc.py      # Please modify `pre_dir_name` and `json_dir_name` in evaluate_acc.py
   

2. Install CULane evaluation tools, please see INSTALL.md

3. generate F, mIoUevaluate_acc after the CULane evaluation tools are installed

   1. all pred txt files will be generated under MMA-Net/evaluation/txt/pred_txt after this step

      python generate_iou_pred_txt.py      # Please modify `pre_dir_name` and `json_path` in  `generate_iou_pred_txt.py`
      

   2. results_MMA and temp_MMA will be generated under MMA-Net/evaluation/txt/results_txt after this step.

      results_MMA: evaluation results of each sequence

      temp_MMA: temporary files generated during evaluation, you can ignore them

      python evaluate_iou.py      # `data_root` should be set as your VIL-100 dataset path in `evaluate_iou.py`
      

   3. Attention!! if you want to evaluation results one more time, please delete all folders/files under MMA-Net/evaluation/txt/results_txt .