Skip to content

Latest commit

 

History

History
99 lines (71 loc) · 3.62 KB

README.md

File metadata and controls

99 lines (71 loc) · 3.62 KB

Spectrogram Generator 📊🎶

This project generates spectrograms from MATLAB .mat files or HDF5 .h5 files. The input files are stored in a directory named input_files, and the generated spectrograms are saved in a directory named spectrograms.

Contents

How to Use 🚀

You can generate spectrograms using a single script.

Running the Script

  1. Place your .mat or .h5 files in the input_files directory.
  2. Run the script to generate spectrograms, specifying the input folder, output folder, range bins, number of pixels for bounding boxes, and optional time slices.

Command

To generate spectrograms:

python gen_spec.py /path/to/input/folder --output_folder /path/to/output/folder --range_bins 0-10 --n_pixels 40 --filter_order 1 --time_slices 32

To boost the color intensity in the spectrograms, add the --boost_colors flag:

python gen_spec.py /path/to/input/folder --output_folder /path/to/output/folder --range_bins 0-10 --n_pixels 40 --filter_order 1 --time_slices 32 --boost_colors

Arguments

  • /path/to/input/folder (required): The path to the directory containing your .mat or .h5 files.
  • --output_folder (optional): The path to the directory where the spectrograms will be saved. If not specified, the script will save the spectrograms in the spectrograms folder.
  • --range_bins (required): Specifies the range of bins to process. You can specify a single bin (e.g., 120) or a range of bins (e.g., 0-10).
  • --n_pixels (optional): The number of pixels around the ground truth frequency for the bounding box. The default value is 40.
  • --filter_order (optional): The order of the high pass filter, if applying a filter.
  • --time_slices (optional): Number of time slices to process (1 to 32). The default value is 1.
  • --boost_colors (optional): Boost the color intensity in the spectrograms. By default, this option is disabled.

Directory Structure 📁

The generated spectrograms will be saved in the following directory structure:

spectrograms/
├── annotations.json
└── drone_name-timestamp-tiltangle-propeller-throttle
    ├── details.txt
    ├── Labeled
    │   ├── spec_labeled_range_bin=x.png
    │   ├── spec_labeled_range_bin=x+1.png
    │   └── ... (other range bins)
    └── Raw
        ├── spec_raw_range_bin=x.png
        ├── spec_raw_range_bin=x+1.png
        └── ... (other range bins)

Organization

To move the raw images into training, testing, and validation splits for model training, run

./organize_imgs.sh

Annotations 📝

The annotations.json file contains the bounding box annotations for all the processed images in COCO format. It is saved in the spectrograms folder and includes the following information for each annotated image:

  • id: The unique identifier for the annotation.
  • image_id: The ID of the image.
  • category_id: The category ID (1 for drone frequency).
  • bbox: The bounding box coordinates [x, y, width, height].
  • area: The area of the bounding box.
  • iscrowd: Specifies whether the annotation is a crowd (always 0 for this case).

Details File 📄

The details.txt file in each subdirectory contains metadata about the spectrograms, including:

  • Drone Name
  • Time Stamp
  • Tilt Angle
  • Propeller
  • Throttle
  • Range Bins
  • Expected Frequency
  • Fill Factor

Enjoy generating your spectrograms! 🎉📈