/comp9517-simPY-round-2

Primary LanguageJupyter Notebook

comp9517-simPY-round-2

Authors

  • Evan Lee
  • Theodore Koveos
  • Ian Ng
  • Rohan Maloney
  • Andrew Timkov

Instructions

Distance and Velocity Estimation

  • Copy velocity.ipynb into Google Colab (will not work on Jupyter Notebook)

  • Setup a folder in your Google Drive with the path Colab Notebooks/comp9517/

  • In this folder ensure there are the following files:

    • condensed_data.zip - a zip folder with all of the velocity training and testing data folders
    • mask_rcnn_coco.h5 - pretrained weights which can be downloaded from here
    • car_mask_rcnn.h5 - pretrained weights fitted for the TuSimple dataset. This is optional as the model training will create this file
    • mrcnn folder - contains source code for Matterport's Mask R-CNN implementation with some edits
    • evaluate folder - contains source code for TuSimple's evaluation code with some edits

  • Run all code cells

  • Execution calls has the structure of make_predictions(TESTING_DIR, detection_mode=["distance", "velocity"], show_img=True, show_all=False, my_evaluate=True, save=True, subset=[10, 23])

    • TESTING_DIR - directory with testing data
    • detection_mode - distance to calculate distance, velocity to calculate velocity
    • show_img - set to True if you want the images to be shown
    • show_all - set to True if you want False positives (un-annotated vehicles) to be shown
    • my_evaluate - set to True if you want each input to be evaluated for accuracy
    • save - set to True if you want images to be saved to a file
    • subset - define a list of clip number that you want to run predictions for. If this is set to [] or ommitted, then predictions will be run for every clip in TESTING_DIR

Lane Detection

  • Open lane_detection.ipynb
  • choose which dataset you wish to run the code on in cell 2
  • choose which clip from the dataset you would like to run the code on