Each file contains data for one PT event (with parent meteor) representing all PTs observed during our campaign.

File names are encoded as: MJD_UTChour_Frame_Label.hdf5
    MJD: Modified Julian Date of recording
    UTChour: The UTC hour when the meteor occurred
    Frame: The frame within the hour dataset
    Label: The unique label given to the meteor in the hour dataset
    
    (Frame/Label have relevance to us, but don't have much significance except to create a unique filename.
    This filename can be matched to GMN data (if available) via the "Event_id" column of the PT_meteors.csv 
    file.)

All data are saved in hdf5 format containing two datasets: "images" and "times".
    "images": the data itself, structured in the order [frame, row, column]. Frames 0 and 1 are pre-meteor
    background images (except for 59633_06_001_0001 and 59730_05_001_0001 which only have one background image).
    Frame 2 is the meteor (occasionally lasting to Frame 3), and Frame 3 onward contain the PT. 
    
    "times": the Unix timestamp corresponding to each frame. Timestamp is recorded after the 5 second 
    exposure + ~1.5 second readout time; to get the time of the beginning of the image, subtract 6.5 from the
    timestamp value.
    
    
The following Python snippet allows the data to be visualized in a way that closely matches how it was 
originally viewed during classification:

    import h5py
    import matplotlib.pyplot as plt
    import matplotlib.animation as animation

    filename = #"insert path to file"#
    im_cube = h5py.File(filename, "r").get("images")[:,:,:].astype(float)

    fig, ax = plt.subplots()
    ax.axis("off")
    ims = []
    for i in range(im_cube.shape[0]):
        frame = ax.imshow(im_cube[i], vmin=50, vmax=300, cmap='viridis', animated=True)
        ims.append([frame])
    ani = animation.ArtistAnimation(fig, ims, interval=50, blit=True, repeat_delay=1000)
    plt.show()