/context-based-parking-slot-detect

Primary LanguagePython

context-based-parking-slot-detect

Tensorflow implementation of Context-based parking slot detection (IEEE Access)

This implementation is based on https://github.com/wizyoung/YOLOv3_TensorFlow

Prepare Dataset (PIL-park)

  1. This code should be run only once at the beginning.

  2. Download Train Dataset

  • link
  • Unzip to $your_data_path/train folder
  1. Download Test Dataset
  • link
  • Unzip to $your_data_path/test folder
  1. Data augmentation, create tfrecord and text files
  • python prepare_data.py --data_path=$your_data_path

Train Dataset

  1. Download pretrain weight (Updated 2020.10.26)
  • link
  • Save to 'pre_weight' folder under "context-based detect" folder

  1. python train.py --data_path=$your_data_path

  2. Trained Weight path

  • Weight files of parking context recognizer are saved to 'weight_pcr/YYYYMMDD_HHMM'
  • Weight files of parking slot detector fine-tuned for parallel parking slots are saved to 'weight_psd/type_0/YYYYMMDD_HHMM'
  • Weight files of parking slot detector fine-tuned for perpendicular parking slots are saved to 'weight_psd/type_1/YYYYMMDD_HHMM'
  • Weight files of parking slot detector fine-tuned for diagonal parking slots are saved to 'weight_psd/type_2/YYYYMMDD_HHMM'

Test Method (with downloaded weight files)

  1. Download trained weight
  • link
  • Unzip under main path (locate "weight_pcr" and "weight_psd" under "context-based detect" folder)
  1. Evaluate
  • python test.py --data_path=$your_test_path

Test Method (with your trained weight files)

  1. Evaluate
  • python test.py --data_path=$your_test_path --pcr_test_weight='weight_pcr/YYYYMMDD_HHMM/cp-0050.ckpt' --psd_test_weight_type0='weight_psd/type_0/YYYYMMDD_HHMM' --psd_test_weight_type1='weight_psd/type_1/YYYYMMDD_HHMM' --psd_test_weight_type2='weight_psd/type_2/YYYYMMDD_HHMM'