DMPR-PS

This is the implementation of DMPR-PS using PyTorch.

Requirements

• PyTorch
• CUDA (optional)
• Other requirements

Pre-trained weights

The pre-trained weights could be used to reproduce the number in the paper.

Inference

• Image inference

(shell)
    python inference.py --mode image --detectorweights $DETECTORWEIGHTS --inference_slot

• Video inference

(shell)
    python inference.py --mode video --detectorweights $DETECTORWEIGHTS --video $VIDEO --inference_slot

Argument DETECTOR_WEIGHTS is the trained weights of detector. Argument VIDEO is path to the video. View config.py for more argument details.

Prepare data

• Download ps2.0 from here, and extract.
• Download the labels, and extract.

• Perform data preparation and augmentation:

(shell)
    python preparedataset.py --dataset trainval --labeldirectory $LABELDIRECTORY --imagedirectory $IMAGEDIRECTORY --outputdirectory $OUTPUT_DIRECTORY
    python preparedataset.py --dataset test --labeldirectory $LABELDIRECTORY --imagedirectory $IMAGEDIRECTORY --outputdirectory $OUTPUT_DIRECTORY

Argument LABEL_DIRECTORY is the directory containing json labels. Argument IMAGE_DIRECTORY is the directory containing jpg images. Argument OUTPUT_DIRECTORY is the directory where output images and labels are. View prepare_dataset.py for more argument details.

Train

(shell)
python train.py --datasetdirectory $TRAINDIRECTORY

Argument TRAIN_DIRECTORY is the train directory generated in data preparation. View config.py for more argument details (batch size, learning rate, etc).

Evaluate

• Evaluate directional marking-point detection

(shell)
    python evaluate.py --datasetdirectory $TESTDIRECTORY --detectorweights $DETECTORWEIGHTS

Argument TEST_DIRECTORY is the test directory generated in data preparation. Argument DETECTOR_WEIGHTS is the trained weights of detector. View config.py for more argument details (batch size, learning rate, etc).

• Evaluate parking-slot detection

(shell)
    python psevaluate.py --labeldirectory $LABELDIRECTORY --imagedirectory $IMAGEDIRECTORY --detectorweights $DETECTOR_WEIGHTS

Argument LABEL_DIRECTORY is the directory containing testing json labels. Argument IMAGE_DIRECTORY is the directory containing testing jpg images. Argument DETECTOR_WEIGHTS is the trained weights of detector. View config.py for more argument details.

Citing DMPR-PS

If you find DMPR-PS useful in your research, please consider citing:

()
@inproceedings{DMPR-PS,
Author = {Junhao Huang and Lin Zhang and Ying Shen and Huijuan Zhang and Shengjie Zhao and Yukai Yang},
Booktitle = {2019 IEEE International Conference on Multimedia and Expo (ICME)},
Title = {{DMPR-PS}: A novel approach for parking-slot detection using directional marking-point regression},
Month = {Jul.},
Year = {2019},
Pages = {212-217}
}