RoadMarkerDetectionUsingSAM

Test for UAV or top down view street image data. Detect basic road marker on road. Implement base on meta segment anything.

Installation

This project depends on PyTorch. Please install it first by following the official instructions for your specific hardware (CPU or GPU with CUDA) on the PyTorch website.

For example, for a recent version of CUDA:

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

Or for CPU only:

pip install torch torchvision torchaudio

Once PyTorch is installed, you can install this project in editable mode:

pip install -e .

Prepare

Download checkpoints sam_vit_h_4b8939.pth from meta segment anything change checkpoints path in config.yml

Config Parameters

• ColorFilter (Keep the color we need as road marker)

  • color_list: The color that needs to be retained as the mask image
  • color_threshold: Threshold determines how close the colors are
  • area_threshold: The noise area size threshold to be removed

• Predictor (meta sam predict)

  • checkpoint: checkpoint path
  • area_threshold: Maximum mask size to be retained

Usage

import RoadMarkerDetectionUsingSAM see pipeline.py for example usage

Or directly use each component:

Predict each element on road python -m predictor.py -i <input_image> -o <output_path>

Test close color filter on road python -m color_filter.py -i <input_image> -o <output_path>