Dongkwon Jin and Chang-Su Kim

Official implementation of "OMR: Occlusion-Aware Memory-Based Refinement for Video Lane Detection".

• Arxiv Paper
• Supplementary Material

• Python >= 3.6
• PyTorch >= 1.10
• CUDA >= 10.0
• CuDNN >= 7.6.5

1. Clone the repository:

   git clone https://github.com/dongkwonjin/OMR.git
   cd OMR

2. Download pre-trained models and preprocessed data:

   • Pre-trained model parameters
   • Preprocessed data

   unzip pretrained.zip
   unzip preprocessing.zip

3. Create and activate a conda environment:

   conda create -n OMR python=3.8 anaconda
   conda activate OMR

4. Install dependencies:

   conda install pytorch==2.0.0 torchvision==0.15.0 torchaudio==2.0.0 pytorch-cuda=11.7 -c pytorch -c nvidia
   pip install -r requirements.txt

   You can also find more options for installing PyTorch here.

1. OpenLane-V: Download OpenLane-V and place it in the original OpenLane dataset directory.

2. VIL-100: Download from here.

3. KINS: Download from here.

ROOT
├── Preprocessing           # Data preprocessing code
│   ├── VIL-100             # Dataset: VIL-100, OpenLane-V
│   │   ├── P00             # Preprocessing step 1
│   │   │   ├── code
│   │   ├── P01             # Preprocessing step 2
│   │   │   ├── code
│   │   └── ...
│   └── ...
├── Modeling                # Model code
│   ├── VIL-100             # Dataset: VIL-100, OpenLane-V
│   │   ├── ILD_cls         # ILD module for predicting lane probability map and latent obstacle mask
│   │   │   ├── code
│   │   ├── ILD_reg         # ILD module for regressing lane coefficient maps
│   │   │   ├── code
│   │   ├── OMR             # OMR module
│   │   │   ├── code
│   ├── OpenLane-V
│   │   ├── ...
├── pretrained              # Pretrained model parameters
│   ├── VIL-100
│   ├── OpenLane-V
│   └── ...
├── preprocessed            # Preprocessed data
│   ├── VIL-100
│   │   ├── P00             
│   │   │   ├── output
│   │   ├── P02             
│   │   │   ├── output
│   └── ...
├── OpenLane                # Dataset directory
│   ├── images
│   ├── lane3d_1000         # Not used
│   ├── OpenLane-V
│   │   ├── label
│   │   ├── list
├── VIL-100
│   ├── JPEGImages
│   ├── Annotations         # Not used
└── ...

To evaluate using VIL-100, follow these steps:

1. Install Evaluation Tools

   • Download the official CULane evaluation tools from here.
   • Save the tools in the ROOT/Modeling/VIL-100/MODEL_NAME/code/evaluation/culane/ directory.

   cd ROOT/Modeling/VIL-100/MODEL_NAME/code/evaluation/culane/
   make

2. Refer to the Installation Guidelines

   • For detailed installation instructions, refer to the installation guideline.

1. Configure Training

   • Set the dataset (DATASET_NAME) and model (MODEL_NAME) you want to train.
   • Specify your dataset path using the -dataset_dir argument.

   cd ROOT/Modeling/DATASET_NAME/MODEL_NAME/code/
   python main.y --run_mode train --pre_dir ROOT/preprocessed/DATASET_NAME/ --dataset_dir /path/to/your/dataset

2. Optional: Edit Configuration

• Modify config.py to adjust the training parameters as needed.

1. Evaluate Pre-Trained Models

   • To get performances of pre-trained models:

   cd ROOT/Modeling/DATASET_NAME/MODEL_NAME/code/
   python main.y --run_mode test_paper --pre_dir ROOT/preprocessed/DATASET_NAME/ --paper_weight_dir ROOT/pretrained/DATASET_NAME/ --dataset_dir /path/to/your/dataset
   

2. Evaluate Your Trained Model

   • To evaluate a model you have trained:

 cd ROOT/Modeling/DATASET_NAME/MODEL_NAME/code/
python main.y --run_mode test --pre_dir ROOT/preprocessed/DATASET_NAME/ --dataset_dir /path/to/your/dataset

3. (Optional) Visualize Results
   • To visualize detection results, set disp_test_result=True in code/options/config.py.

Preprocessing data involves several steps:

1. Convert Ground-Truth Lanes
   • Convert ground-truth lanes to pickle format (VIL-100 specific).
2. 2D Point Representation
   • Represent each lane in the training set as 2D points sampled uniformly in the vertical direction.
3. Lane Matrix Construction
   • Construct a lane matrix, perform SVD, and transform each lane into its coefficient vector.
4. Generate Video-Based Datalists
   • Create datalists for training and test sets.

 cd ROOT/Modeling/DATASET_NAME/MODEL_NAME/code/
python main.y --run_mode test --pre_dir ROOT/preprocessed/DATASET_NAME/ --dataset_dir /path/to/your/dataset

@Inproceedings{
    Jin2024omr,
    title={OMR: Occlusion-Aware Memory-Based Refinement for Video Lane Detection},
    author={Jin, Dongkwon and Kim, Chang-Su},
    booktitle={ECCV},
    year={2024}
}